Rahul Raj
Share
  •  
  •  
  •  
  •  

When veteran filmmaker Garry Marshall showed ‘The Princess Diaries’ to his five year old granddaughter, she was not happy with its ending. She wanted to see the gorgeous castle that Princess Mia would live in. The director then had to reshoot the ending to include a glimpse of the castle to pacify his granddaughter. Gathering customer feedback is not this straightforward in the real world – even though customers are more vocal about their opinions now than ever before.

There is no dearth of data: retailers such as Amazon and Flipkart encourage customers to post reviews on product purchases. But sifting through all this data is a tedious process. Businesses wanting to design strategies based on customer sentiment find it difficult to make sense of the vast amounts of information available. Imagine if Garry Marshall had millions of grandkids, each with a different opinion!

 Statistical Algorithms to the Rescue

Customer reviews drive sales. According to Nielsen’s Global Trust in Advertising Report for 2015, 66% of global consumers indicated they trust or act upon online reviews. The objective of sentiment analysis is to convert the subjective information available online in the form of customer reviews into meaningful information that can help with decision-making. If your goal is to generalize customer reviews and compute an overall rating for a product or service, the Pointwise Mutual Information and Information Retrieval (PMI-IR) method is an effective choice. It takes out the adjectives from the reviews and estimates the semantic orientation. PMI-IR classifies reviews according to the average semantic orientation, and is very useful in summarizing online reviews. marketer can use this to track the general opinion on a new ad campaign.

There are various statistical algorithms that can help with sentiment analysis. Machine learning algorithms such as Support Vector Machines (SVM) are useful classification techniques. SVMs define a clear boundary between positive and negative sentiment categories, and map customer reviews into each category. The objective of SVM is to locate a decision boundary that divides the data in a way that maximizes the margin for greater accuracy. The margin is the region between the data classes on either side of the boundary. This technique works well for classifying traditional text.

A key challenge in analyzing customer reviews is to understand the sentiment behind the reviews with respect to individual product attributes or features. This is because a positive review does not mean that the customer liked everything about the product or service. Similarly, a negative review does not imply that the customer did not find any likeable feature in the product. So, it helps to classify positive and negative sentiments, not just at the overall product level, but also at the product attribute and feature level. A ranking methodology based on product features can be used to mine customer sentiment here.

This method selects an array of features important to customers and applies text mining techniques to the reviews to build a feature specific graph. The product is then ranked by mining the graph and using a page ranking algorithm. For example, the battery life of a laptop maybe good, but it may be lacking in visual appeal, resulting in a lower rank for the latter. By using an approach that considers the ranks and weights, the laptop manufacturer can figure out which aspects of the product to focus on for making further improvements.

To assign these ranks, text mining techniques that use linguistics processing methods are used. Text has layers of hidden emotions such as sarcasm and irony. For example, a negative review may end with a sarcastic “what a great product!” The algorithm should be able to identify these emotions and finally classify a review as positive, negative, or neutral by understanding the ways in which context can affect sentiment.

Translating Customer Sentiment into Business Value

Customers now routinely look up products online before purchasing them. Technology and social media have made it easier for customers to make purchase decisions by enabling product reviews and peer discussions. For organizations, this offers a gold mine of information on the customer’s lifestyle, hobbies, purchasing preferences, and other aspects.

That brings us to the important question – how does a business identify the right technique to use? Start by identifying your objective: Do you want to improve sales conversion? Enhance product quality? Reduce customer attrition? Once the objective is clear, it becomes easier to zero in on the right data source (feedback from social, reviews, and surveys) and identify the best suited analytical tools and techniques to maximize value.

Which analytical tools and techniques do you find the most reliable and relevant to your business? Share your thoughts in the comments section below.


Share
  •  
  •  
  •  
  •