Project

      The purpose of the project is to use semantic information from financial news to predict changes in stock prices of various companies.  Extracting this kind of information is very complex and requires many steps, including locating company names from the text, determining the relevant semantic information, and constructing a model to generate accurate predictions.  I worked on improving the overall performance by focusing on capturing additional company mentions that were missed.  To improve this named entity recognition, I manually inspected news articles and observed instances of company names that were missed in order to modify the algorithm.  Furthermore, entities can be referred to in other ways, such as through the use of pronouns or nominal phrases.  The instances that all refer to the same entity are called coreferences and the process of capturing the coreferent entities is called coreference resolution.  Coreference resolution was not included in the original program, but I incorporated it into the modified version that I worked on.  Modifying the original named entity recognition as well as incorporating coreference resolution resulted in a significant increase in the number of company mentions captured.  However, the overall prediction score did not increase, but I speculated reasons as to why that may have been the case.  I learned a great deal about this project and thoroughly enjoyed this DREU experience!