An Inheritance-Based Lexical Approach to Sentiment Analysis [Problem Statement]

 

.

1.4 Challenges

Sentiment analysis is a popular research area, and there have been over 7000 research projects and articles written on the topic. Nevertheless, there are still many major challenges, some of which have been identified at previous years’ Sentiment Analysis Symposium, an annual conference that addresses the business value of sentiment, opinion, emotion and intent, including:

1. There is a lack of suitable modelling of compositional sentiment, which means that the overall sentence sentiment of the sentiment bearing word, the sentiment shifters and the sentence structure need to be calculated more accurately at the sentence level.

2. Sentiment lexicons are one of the important features used in sentiment analysis. Creating a sentiment lexicon is another challenging task. Building a lexicon with semantic intensity scores is extremely beneficial. However, having such scores annotated by human annotators is not feasible as it is difficult to maintain consistency across different annotators. Various lexicon sources are publicly available for sentiment analysis. However, which sources give the most reliable semantic scores has not yet been established.

3. In the same document, a product may be referred to by many names. This is one of the main issues of automatic name entity resolution, and has not yet been solved effectively. The handling of anaphora resolution in an accurate way is another major issue in text mining. It is an important, challenging issue in sentiment analysis too.

4. It is essential to identify the text relevant to each entity, when there are several entities discussed in a document. The current accuracy of the identification of relevant text does not give satisfying results.

5. Another big challenge in sentiment analysis system is handling noisy text (text with spelling/grammatical mistakes, missing/problematic punctuation, slang, etc.).

6. Handling sarcasm and irony is another challenge in sentiment analysis. Some reviews tend to show their dissatisfaction towards a product/service in a sarcastic way using positive language. Identifying sarcasm has not yet been properly integrated within sentiment analysis systems, although some previous researchers (e.g. (Riloff et al., 2013)) have worked on recognition of sarcasm in the field.

7. A new approach is needed to handle factual statements. Many statements about factual entities carry sentiment. But only subjective statements are considered in most of the current sentiment analysis methods, and researchers fail to consider such factual (objective) statements.

8. Some authors like to use ambiguous comments in their posts. Ambiguous words and statement may be humorous but can lead to vagueness and confusion. Expressing the meaning of such statements without context is difficult.

9. In some cases, applications translate foreign customers’ reviews into English. Many translation programs have difficulty correctly interpreting sentiments in language, as Western and Asian or African sentiments differ from each other significantly

.