Searching for Diverse Perspectives in News Articles: Using an LSTM Network to Classify Sentiment

When searching for emerging news on named entities, many users wish to find articles containing a variety of perspectives. Advances in sentiment analysis, particularly by tools that use Recurrent Neural Networks (RNNs), have made impressive gains in their accuracy handling NLP tasks such as sentiment analysis. Here we describe and implement a special type of RNN called a Long Short Term Memory (LSTM) network to detect and classify sentiment in a collection of news articles. Using an interactive query interface created expressly for this purpose, we conduct an empirical study in which we ask users to classify sentiment on named entities in articles and then we compare these sentiment classifications with those obtained from our LSTM network. We compare this sentiment in articles that mention the named entity in a collection of news articles. Last, we discuss how this analysis can identify outliers and help detect fake news articles.