Web search model for Automatic Annotation of Tweets in Opinion Mining

Abstract

In sentiment analysis, annotation is the crucial step of data processing to label the review or sentences as positive, negative, or neutral. An annotation process is usually performed by three key approaches: (i) manual, (ii) crowdsourcing, and (iii) automated annotation. Manual annotation is preferred in most of the literature’s and crowdsourcing tools are used in some of the works. This indicates that there is a scarce of automatic annotation and its service is highly essential to support more systematic research in sentiment analysis. Manual procedures mostly depends on external annotators, resulting in costly and time-consuming processes. Thus, we propose a method for automatic annotation using web search model to dynamically label reviews (positive, negative, or neutral) that are not available in the dictionary. Some research works consider the product opinions from e-commerce sites where most of the texts may not be available in the dictionary, in such cases, the web search model can be used instead of manual annotation. A large-scale opinosis dataset is used to evaluate the accuracy of the algorithms and feasibility of the model. The experimental results indicate that this model outperforms conventional methodologies and therefore we firmly believe it will be useful for current researchers in the field of opinion mining.