Ranking Linked-Entities in a Sentiment Graph

Reputation analysis is naturally associated to a sentiment analysis task of the targeted named-entities. This analysis leverages on a sentiment lexicon that includes general sentiment words that characterize the general sentiment towards the targeted named-entity. However, in most cases, target entities are themselves part of the sentiment lexicon, creating a loop from which it is difficult to infer an entity reputation. Sometimes, the entity became a reference in the domain and is vastly cited as an example of a highly reputable entity. For example, in the movies domain it is not uncommon to see reviews citing Batman or Anthony Hopkins as esteemed references. In this paper we describe a three-step procedure to perform reputation analysis of linked entities. First, our method jointly extracts named entities reputation and a domain specific sentiment lexicon. Second, an entities graph is created by analyzing cross-citations in subjective sentences. Third, the entities reputation are updated through an iterative optimization that exploits the graph of the linked-entities. The proposed approach closely models real-world domains, where domain specific jargon is common and entities are so popular that they become widely used as sentiment references. The evaluation on a graph with 12,687 vertices, of which 3,177 are linked entities and 9,510 are sentiment words, shows that our approach can improve the correct detection of an entity's reputation.