Cross-Domain Knowledge Adaptation Using Iterative Scaling and Language Model

Automatic opinion mining of social media textual data is crucial for determining people's attitudes and improving decision making. Since most of the reviews are unstructured, they require processing steps such as clustering or classification to extract important information. Opinion expression varies in every other domain and annotating data for all the possible entities is both expensive and timeconsuming. Classifiers when trained on a specific domain don't perform poorly when we apply them directly to other domains. The domain adaptation's task is to extract the features and appropriate opinionated words from any specific domain and analyze the sentiment on other domains after mapping. To concern this issue, we propose a framework to share domain invariant knowledge of two different domains. To extract significant information, the context and co-relation between features are analyzed. Domain-independent and specific features are subsequently separated from these features depending on their tf-idf values. When independent words are used as a bridge, clustering technique is used for aligning domain-specific features. Additionally, the features which are used to train a classifier can also predict opinion words for the target domain. A series of trials demonstrated that the suggested framework achieved <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$81.05 \%$</tex> which is comparable with the baseline methods.