Evaluation of Sentiment Databases
Marc Notz, Jens Grambau, Arno Hitzges · 2019
Nowadays it happens more and more frequently that people get their opinions about products in social media or online reviews. For this purpose, product reviews themselves or channels on social media are searched in order to get an impression of what other users think about products before buying. For companies, these Social Listening Statements (SLS) can be business critical. A sentiment analysis is part of Natural Language Processing (NLP). The analysis of text tries to identify and extract sentiments in a text. In this paper, two supervised machine learning algorithms for text analysis were combined with four sentiment databases to increase accuracy in recognition. For better comparability, basic sentiment databases without subject-specific terms were used. Experiments were conducted with manually and automatically generated data sets, ranging in size from 300 to 16,000 entries. The combination of these two methods resulted in an accuracy of up to 94% in detecting positive or negative sentiment in texts. The combination of sentiment databases (SD) and machine learning significantly increased the recognition.