Event Detection And Extraction For Financial News Texts

Information extraction,a hot field in natural language processing,aims to extract information of interest from unstructured text and store it in structured form.Event extraction plays an important role in information extraction,event extraction in financial field focuses on extracting hot financial events from unstructured financial news and expresses them with structured form.Traditional event extraction can only extract events of specified types,each event type need a predefined event template and a certain number of annotated corpus.However,the event types we pay attention to in financial field are wide-ranging,from politics to sports and even natural disasters.Therefore,it is impossible to know the specific event types in advance,which challenges the traditional method of event extraction.So we decompose the task into event detection based on text clustering and event extraction based on event clusters,propose an unsupervised framework which achieve accurate and open event extraction.Our main contributions are summarized as follows:(1)In order to detect hot financial events,we propose an event detection algorithm based on text clustering,which designs a feature extraction method to extract event level feature from news to achieve event granularity clustering.We manually constructed the annotated data set to verify the validity of the method.The experimental results show our method can achieve 92% accuracy.(2)After hot financial event detection,we propose an event extraction algorithm based on probability model.The algorithm mainly uses the common semantic analysis technology at the sentence level and the rich redundancy of cluster text to realize unsupervised event extraction.The experimental results show that the proposed method can achieve 73%accuracy.(3)Based on the event detection and extraction algorithm proposed in this paper,a hot event extraction system for financial news is designed and implemented to obtain the structured information of hot financial events online.In general,this thesis proposes an unsupervised hot event detection and extraction algorithm for financial news texts.Based on this algorithm,a real-time financial event extraction system is designed and implemented to discover hot financial events and obtain structured representations of these events.