Construction Of Commonsense Causal Knowledge Base

Commonsense causal reasoning is a central challenge in artificial intelligence.It plays a critical role in people's daily behavior and decision-making.It is the process of capturing and understanding the causal dependencies amongst events and actions.Such events and actions can be expressed in terms,phrases or sentences in natural language text.Causal relations are usually composed of cause and effect part,which are represented as text spans or segaments in corpus.Much interesting information appears as natural language text,which must be processed and analyzed to derive valuable knowledge.Commonsense is one thing that shared by nearly all people and can be used for reasoning task.However,commonsense causal relations in text are sparse,ambiguous,and sometimes implicit,and thus difficult to obtain.This paper aims at better understanding on causal relations which includes the detection,recognition and extracition of causality.We also hope that we can tell whether two envents have potential causal relation.We are going to build two different causal knowledge base to help solving causal reasoning tasks.First one is to build a network of causal-effect terms from a large web corpus named CausalNet.We use traditional frequency and conditional probability as the key technique.In CausalNet,nodes are words and are connected by directed lines with weights on it.Weight of the line measuers the amount of causal strength supposing the start node as cause and end node as effect.Second one is to build word embeddings for words of different roles(cause or effect),which we called CausalVec.We name the space as cause-effect space.Every word has two vector representation corresponding to cause and effect roles.We measuer the amount of causal strength by calculating the dot product of two vectors.With causal knowledge base,we are able to do causal reasoning and it becomes easy to track the most possible effect word given cause word and vice versa.