Abstract
User-generated content provides managers rich data opportunities to gauge customers’ feelings, and sentiment analysis of social media data is increasingly used in a wide variety of fields, including finance, politics, and marketing. Popular tools include simple volume measures (such as number of likes and clicks), dictionary-based analysis of sentiment-rich words, and machine learning-based techniques.
These tools are typically used in isolation, which hinders ability to compare their power to explain customer-level performance such as brand awareness, image, purchase intent, customer satisfaction, and recommendation. Similarly, managers and other decision makers are uncertain as to whether their analysts are using the appropriate tool to extract actionable sentiment information for their specific brands and industries.
Raoul Kübler, Anatoli Colicev, and Koen Pauwels address this issue by analyzing a unique dataset with daily social media and consumer mindset metrics for 48 brands in diverse industries. They compare the dynamic explanatory power of these methods for daily customer mindset metrics.
Overall, vector autoregressive models show that machine learning-based approaches have the best average performance for detecting positive, negative and neutral comments. At the same time, there are systematic differences among industries, mindset metrics, and brands. Since costs increase when moving from volume metrics to dictionaries and machine learning, these findings can help managers decide when more detailed analytics are worth the investment.
Volume metrics are the cheapest to obtain and perform well for explaining awareness and purchase intent. This is an appropriate choice especially for smaller brands in negative sentiment industries (e.g., airlines and banking). Managers of other brands may consider investing in more sophisticated sentiment analysis techniques, especially when explaining or predicting brand impression, satisfaction, and recommendation.
These tools are typically used in isolation, which hinders ability to compare their power to explain customer-level performance such as brand awareness, image, purchase intent, customer satisfaction, and recommendation. Similarly, managers and other decision makers are uncertain as to whether their analysts are using the appropriate tool to extract actionable sentiment information for their specific brands and industries.
Raoul Kübler, Anatoli Colicev, and Koen Pauwels address this issue by analyzing a unique dataset with daily social media and consumer mindset metrics for 48 brands in diverse industries. They compare the dynamic explanatory power of these methods for daily customer mindset metrics.
Overall, vector autoregressive models show that machine learning-based approaches have the best average performance for detecting positive, negative and neutral comments. At the same time, there are systematic differences among industries, mindset metrics, and brands. Since costs increase when moving from volume metrics to dictionaries and machine learning, these findings can help managers decide when more detailed analytics are worth the investment.
Volume metrics are the cheapest to obtain and perform well for explaining awareness and purchase intent. This is an appropriate choice especially for smaller brands in negative sentiment industries (e.g., airlines and banking). Managers of other brands may consider investing in more sophisticated sentiment analysis techniques, especially when explaining or predicting brand impression, satisfaction, and recommendation.
| Original language | English |
|---|---|
| Journal | MSI Working Paper Series |
| Publication status | Published - 2017 |