| Summary: | INTRODUCTION: The requirement of the labeled dataset in the source domain makes the Cross Domain Sentiment Classification (CDSC) task complicate in the situation when the dataset is labeled manually. OBJECTIVES: To overcome the dependency of CDSC tasks on manual labeling of the dataset by proposing a polarity detection task. METHODS: We have proposed the CDSC-PDT method that is the polarity Detection Task (PDT) followed by the CDSC task. The proposed PDT task extracts the polarity of reviews from the source domain using the contextual and relevancy information of words in documents and this automatic labeled dataset is further used to train classifiers to make the further classification. RESULTS: Proposed method is comparable to the traditional learning method giving the highest precision 85.7%. CONCLUSION: The proposed method does not need to manually label the documents in either of the domain (source or target), hence it overcomes the human intervention and is also time saving and cheap process, unlike traditional CDSC tasks.
|
|---|
