Converging Human Knowledge for Opinion Mining

Opinion mining focuses on analyzing opinions in documents. Existing most algorithms for mining opinion either are machine-only, leaving plenty of confused puzzles due to lacking human background knowledge, or using opinion dictionary from domain experts. The latter is expensive and hard to scale. In this paper, we propose a novel approach RULING (conveRging hUman knowLedge opInion miNinG) for opinion mining, where human include both the crowd and the experts. Firstly, we propose a method for combining expert knowledge with the machine learning method. Then we use the prediction result to find out the hard item, and classify them using crowdsourcing. This method can scale better than the previous methods and get a better result. Experimental results demonstrate our RULING approach outperforms related proposals in terms of classification performance.