Sentiment Analysis Approach for Reputation Evaluation

Link to slide: SlideServe

Research Question:

• Can computers perform reputation evaluation that benefit businesses?
• How sentiment analysis can be used to evaluate reputation of a product or services?

Objective

• Evaluate and Compare the selected sentiment classification techniques used to evaluate brand reviews.
• Findings are presented to make informative decisions regarding the adoption of classification techniques.

Proposed Methodology

• Dataset containing 400,000 reviews of unlocked mobile phones sold on Amazon was selected.
• Three approaches (lexicon-based, machine learning and hybrid) were implemented to identify the underlying sentiment.
• Model evaluation metrics were utilised for comparative analysis.

Programming Language and Software Tools

• PyCharm by JetBrains
• Python
• Scikit-Learn Library

Results

• Accuracy of Hybrid approach was the highest, giving 81.2% of correctly predicted observation.
• Precision score of Lexicon-based approach was the lowest with 54.0% of correctly predicted positive observations.
• F1 score of Hybrid approach was the highest, presenting with 70.2% of harmonic mean between precision and recall.
• The positive sentiment label in Apple and BlackBerry mobile reviews were higher, compared to the negative and neutral sentiment labels.

Conclusions

• Hybrid approach to sentiment analysis can effectively be used to evaluate brand reviews that benefit businesses.
• Underlying sentiment of brand reviews can be evaluated with the use of sentiment classification techniques.
• Slang and emoticons handling may be implemented to improve results of sentiment analysis.