Sentiment Analysis Method

Preparation of data - including basic preprocessing steps like removing the markup tags and content which is not-textual, as well as unnecessary data, such as dates of the texts and names of the authors. \item Text analysis - the analysis of text linguistic features, usually during this step POS tagging and negation tagging are conducted. \item Sentiment classification itself and results of this classification. \end{enumerate}

Concerning the sentence-level, \citet{27} formulate the sentence-level sentiment classification task as a \lq\lq sequence labelling problem\rq\rq. The data used by their model are sentence-segmented documents divided into sentences annotated with sentence-level sentiment labels (positive, negative or neutral). \\

One of the main challenges during the process of sentiment analysis is that a word in a sentence can have different connotations according to the context, thus, sentiment analysis has to take the context into account. For example, the word \textit{cheap} is positive if used by the buyer in the sphere of commerce, but it is negative if it is used to describe someone's behaviour in conversational speech (when a person shows a lack of honesty and moral principles). Moreover, the contextual polarity changes according to the person who is speaking about the sentiment. For instance, consider the sentence in

Many scientists like \citet{55} and \citet{56} referred to the issues of word-level sentiment analysis. The first application of the WordNet to sentiment analysis was conducted by \citet{57} and \citet{58}. They propose the extension to lists of manually tagged positive and negative words with the addition of the synonyms of these words to the list. Then, sentiment polarity was assigned to each word, and depending on the strength of this polarity, the words were ranked. It is known that word sentiments are good indicators of semantic characteristics of a phrase or a text as it was also suggested by \citet{59}. The study on word-level sentiment annotation like in \citet{60} produced a variety of lists of words that were manually or automatically tagged as sentiments. \\