Explaining underlying causes or effects about events is a challenging butvaluable task. We define a novel problem of generating explanations of a timeseries event by (1) searching cause and effect relationships of the time serieswith textual data and (2) constructing a connecting chain between them togenerate an explanation. To detect causal features from text, we propose anovel method based on the Granger causality of time series between featuresextracted from text such as N-grams, topics, sentiments, and their composition.The generation of the sequence of causal entities requires a commonsensecausative knowledge base with efficient reasoning. To ensure goodinterpretability and appropriate lexical usage we combine symbolic and neuralrepresentations, using a neural reasoning algorithm trained on commonsensecausal tuples to predict the next cause step. Our quantitative and humananalysis show empirical evidence that our method successfully extractsmeaningful causality relationships between time series with textual featuresand generates appropriate explanation between them.
展开▼