An E-Commerce Company

CASE STUDY 02

Developing Sentiment Analyzer to track user sentiments

Problem

Solution

Outcome

CASE STUDY 02

The Problem

Sentiment Analysis holds a vital role in performing marketing analysis. Our client wanted to automate the process of reading the feedback left from the customers to determine the user sentiment at consumer level.

The existing process included manual labour to comb through all the customer feedback reviews and present the findings to the Product teams. It was practically impossible to go through tens of thousands of reviews left by the customers. As a result, a handful of the comments were chosen randomly to assess the sentiments behind them.

However, as a data-centric company, our client wanted to utilise all the feedback available to increase statistical significance of the finding and also remove the need of wasting hours of a Marketing team time to do the work manually.

CASE STUDY 02

Solution

For the following weeks, we had to sit for several meetings with the client’s various teams to understand what sort of data were being streamed, how they needed to be processed, and what type of reports the client wanted us to build.

Bleu Leaf Analytics opted for a Machine Learning based solution to analyse sentiments of the customer feedbacks. The feedback was used as raw input for the analyzer. The output represents the percentage of positivity, negativity and neutral sentiments in the given inputs. As the input dataset comprises highly unstructured text-based data, we needed to preprocess it to break it down to a machine friendly format. We used Pandas dataframes to handle the heavy processing. In the preprocessing steps we took care of stopwords, unwanted characters, broken sentences to words, and then each word to their respective root forms. After that we used word vectorizations to transform words to multi dimensional machine readable vectors. Then we calculated the polarity of each word to determine the overall sentiment of the feedback.

After the analysis was done, we generated certain plots to bring out more hidden insights that the feedback had to offer. Among those plots includes word cloud, word frequency chart, bigram and trigram etc.

CASE STUDY 02

Outcome

Our client now can analyse thousands of comments effectively and hassle free. It enabled the Marketing teams to pinpoint areas of interest where development is needed. Also the Product Team could now perform targeted customer segmentation to decide for which customer segment they should introduce new features for their products.
With the new:

N

Track Customer Sentiment throughout the entire customer journey

The Teams now could monitor customer sentiment and trace that data back to the actual event. The teams now had a complete journey of how the users viewed their product.

N

Identify Customer Segments Have the Strongest Opinions

With sentiment analysis our Client could now collect strong positive and negative sentiment. By quickly identifying these people, the marketing could now use that data to create strong brand ambassadors..

N

Improve Product features

With clear feedback from the customers and the sentiment analysis over the time, the Product was well positioned to iterate the product and the brand directly towards the customers.

N

Customer Service Issues prioritisation

Customer support could now be priritoese based on the degree of sentiment from the client. Clients with very strong negative feedback about a product or service could now be placed in front of the queue to have the issue addressed.