Research on Domain-independent Opinion Target Extraction

Opinion Target Extraction is one of the important tasks for text sentiment analysis, which has attracted much attention from many researchers. For this task, we proposed an M-Score algorithm utilized in the model which realized the domain-independent opinion target extraction function. This algorithm is derived from the Pointwise Mutual Information algorithm, but the difference is that it doesn’t need any manual seeds collection or any web searching engines, which reduces the manual participation and easy to be transplanted. This model starts with document preprocessing, effective opinion sentences extraction and candidate opinion target extraction by employing Conditional Random Fields Model with feature templates. Next, the M-Score algorithm is employed to extract seed set, and the bootstrapping approach is invoked to process the candidate opinion targets. Finally, the model uses word frequency and the Noun pruning algorithm to filter the opinion targets, and then obtains the final opinion targets for output. The experimental results show that the M-score method performs better than Pointwise Mutual Information algorithm in precision and recall.