Abstract: The alarm log contains time,space,and case description information,which is unstructured spatiotemporal data.Compared with spatio-temporal social media,there are fewer connections between the data items in the alarm log,and complex network relationships cannot be formed between the data items.It is difficult to provide valuable clues in data patterns.Therefore,its analysis depends more on the semantic mining and the exploration of semantic spatiotemporal patterns.In response to this problem,a visual analysis framework is proposed to support large-scale unstructured alarm log spatiotemporal patterns interactive exploration.First,a method based on topic model integration to extract topics from heterogeneous texts is proposed;Second,the framework includes a data cube to quickly respond to user queries;Third,a visual interactive system is designed and implemented to support the visual interactive exploration of the data cube.Finally,we use a real alarm log of a city in China to evaluate the approach.The rich detected patterns and high topic prediction accuracy prove the effectiveness of the approach.