SYSTEM FOR ENTITY-WISE SENTIMENT CLASSIFICATION USING UNSUPERVISED LEARNING ALGORITHMS

摘要

The present invention relates to unsupervised rule-based textual analysis process embodied in a computer readable machine. This process is not context biased and no training data or supervised learning is required to improve its accuracy. Pre-processing of text data is carried out first wherein transformations are followed by grammar and spelling correction followed by sentence parsing. The sentences are then semantically structured with dependencies and POS tags to make sense from the sentence. Entities are recognized and subject-verb-object-other attribute tuples are formed arising from a sentence. Sentiment weights for each word, dependency structure, and POS-tag based rules are together used to extract sentiment, class, or context from the tuple. This application provides 75-95% accuracy depending on the complexity of the sentence. In real world, 80% of sentences are not complex and this process shows an average accuracy of 83-89%.