Sentiment Analysis - Why You should Do it and How you can do it
5 likes9,803 views
The document discusses the importance of sentiment analysis for improving customer experience, product enhancement, and gaining competitive intelligence, particularly in the tourism industry. Dr. Amancio Bouza outlines a machine learning approach to analyze customer feedback, detailing the process and challenges of handling large volumes of data. The proof-of-concept achieved a sentiment analysis accuracy of 79%, emphasizing its relevance for understanding customer sentiments and enhancing business strategies.
Sentiment Analysis - Why You should Do it and How you can do it
- 1. SENTIMENT ANALYSIS Why and How Swiss Group for Artificial Intelligence and Cognitive Science - 2nd SGAICO at HSLU – May 23rd, 2017 https://sgaico.swissinformatics.org/
- 5. Simple Alternative: Net Promoter Score (NPS) How likely do you recommend Travelito to your friends? 5 very unlikely very likely
- 7. 7 1. Improve Products 2. Enable Closed-Loop Customer Experience 3. Understand Competition
- 9. 9 “The bed in the City Hotel was not comfortable. However, the breakfast was excellent with good food and excellent service. We had a wonderful time and will travel with Travelito again.” Example of customer feedback for anonymized company in tourism industry
- 10. Analyze Entities Related to Negative Feedbacks and Drill Down 10 Named Entity Recognition: - Organization - Location - Person Can you spot the problem? ;)
- 27. 27
- 29. 29 Common Text Preprocessing applied - Lemmatization - Normalize Lower Case - and much more Best Performer for this dataset Two-Class Boosted Decision Tree 500 Features were selected for this dataset selected with 𝜒2 Extraction of N-Gram Features from Text
- 30. Key Experiences with Azure ML Studio for Sentiment Analysis + 1-Gram Features provided best results + Preprocessing and Normalization of Text is simple - Data Preparation was cumbersome with built-in function blocks -> use R - Key Phrase Extraction was useless to extract sentiments - Handover of ML Workspace to others is complicated - Automated creation of Web Service isn’t intuitive. Requires rethinking of the ML workflow + However, it provides great API Documentation, Sandbox and examples on the fly + Intuitive Machine Learning Studio + Fast & simple setup of (scientifically valid) experiments + High Computational Performance + Support for R / R required for advanced stuff, data preparation in particular 30
- 32. 32 “The bed in the City Hotel was not comfortable. However, the breakfast was excellent with good food and excellent service. We had a wonderful time and will travel with Travelito again.” Example of customer feedback for anonymized company in tourism industry
- 34. Evaluation results in detail: 79% Accuracy 34 Accuracy Precision Recall F-Score AUC 0.788 0.751 0.733 0.742 0.853