Causal Relation Extraction And Reasoning From Text

The central aim of many studies is the elucidation of causal relations among variables or events.Conducting controlled experiments,the golden standard method for causal discovery,is too costly in scientific research.Causal knowledge can also be derived from purely observational data.Many studies have been dedicated to extracting causal relations from structured data.Unfortunately,these studies utilize small data or synthetic data,which greatly limits the scalability of their findings for use in real and massive data.Alternatively,we can also discover causal relations in text data like scientific literature,textbook,news,patients' textual records,etc.The research of causal relation analysis from text data consists of three major categories.The first causal relation extraction from text data,which aims at extracting causal relations from pair of entities or pairs of events in text.The second is causal relation reasoning from text data,which aims at 1)generalizing specific causal relations for high-level causal rules,and 2)inferring unseen causal relations and generating causal hypotheses.The third is application based on causal relations,which aims at benefiting prediction and decision-making by making use of causal relations.Our research covers the above three directions.Firstly,we analyze the effective method and features for causal relation extraction from text data.Secondly,from the extracted causal relations,1)we apply inductive inference to get high-level abstract causal relations;2)we apply causal chain of reasoning to infer unseen causal relation and generate new hypotheses.Thirdly,we analyze challenges of using causal relations and propose a method to deal wit.The main contents of our research can be summarized as follows:1.Event causality extraction based on connectives analysis.Causality is an important type of relation which is crucial in numerous tasks.Therefore,causality extraction is a fundamental task in text mining.This paper presents a new Restricted Hidden Naive Bayes model to extract causality from texts.Besides some commonly used features,such as contextual features,syntactic features,position features,we also utilize a new category feature of causal connectives.This new feature is obtained from the tree kennel similarity of sentences with connectives.In previous studies,the features were assumed to be independent,which is not the reality.The advantage of our model lies in its ability to cope with partial interactions among features.Experimental on a public dataset shows that our method goes beyond all the baselines and features of causal connective is the most effective for causal relation extraction.2.Constructing abstract event causality networks via inductive inference.In this study,we formally define the problem of representing and leveraging abstract event causality to power downstream applications.We propose a novel solution to this problem,which build an abstract causality network and embed the causality network into a continuous vector space.The abstract causality network is generalized from a specific one,with abstract event nodes represented by frequently co-occurring word pairs.To perform the embedding task,we design a dual cause-effect transition model.3.Causal relation discovery and hypothesis generation via causal chain of reasoning.This study is to discover unseen casual relations and generate new causal relation hypotheses from medical text data.Existing studies have utilized extraction models to find pseudo causal relation from single sentences,while the knowledge created by causation transitivity –often spanning multiple sentences –has not been considered.Furthermore,we observe that many pseudo causal relations follow the rule of causation transitivity,which makes it possible to discover unseen casual relations and generate new causal relation hypotheses.In this paper,we address these two issues by proposing a factor graph model to incorporate three clues to discover causation expressions in the text data.4.Disease prediction based on causal relations.In order to leverage sparse casual evidence for prediction,this study propose a representation learning method of causal knowledge network with constraints.This method makes it possible to take as input narrative textual symptoms of a patient for disease prediction.In summary,this paper explores characteristics of textual causality for analyzing presentation,generalization,inference and prediction.This research has achieved so preliminary results on causal relation extraction,abstract event causality network construction,new causal hypotheses generation and disease prediction.In the future,we expect that these research results can play a role in giving some inspirations to other researchers,stimulate the enthusiasm of studying textual causality,and hope that relevant practitioners can make use of the practical foundation of this thesis to promote applications in related industries.