Comparison of Data Models for Unsupervised Twitter Sentiment Analysis

"Identifying the sentiment of collected tweets has become a challenging and interesting task. In addition, mining defining relevant features that can improve quality classification system is crucial. The data modeling phase fundamental for whole process since it reveal hidden information from textual inputs. Two models are defined in presented paper, considering Twitter-specific concepts: hashtagbased representation text-based one. These will be compared integrated into an unsupervised determines groups based on labels (positive negative). Moreover, wordembedding techniques (TF-IDF frequency vectors) used to convert representations numeric input needed clustering methods. experimental results show good values Silhouette Davies-Bouldin measures environment. A detailed investigation several items (dataset, method, representation, or word embeddings) checking best setup increasing detecting Twitter’s messages. analysis conclusions first considered more complex experiments. Keywords: Sentiment Analysis, Twitter, Data Representation, Hashtags, Clustering. "