Community Spring Cleaning week is here! Join your fellow Maveryx in digging through your old posts and marking comments on them as solved. Learn more here!

Maveryx Success Stories

Learn how Alteryx customers transform their organizations using data and analytics.
STORIES WANTED

Showcase your achievements in the Maveryx Community by submitting a Success Story now!

SUBMISSION INSTRUCTIONS

Visual Science Completes Sentiment Analysis in Only 3 Days

AlteryxAdvocacy
Alteryx
Alteryx
FullColor_1024x1024_300dpi.jpg

Overview of Use Case

Priscila Castro had been using Alteryx for six years when she decided to open her own consulting firm. Visual Science offers analytics and BI solutions with Alteryx Designer. The company recently acquired Alteryx Intelligence Suite to gain productivity and to scale the rapid creation of predictive models with machine learning, without writing code or running complex statistics. In this use case, Priscila shares how she created a text sentiment analysis for a consumer complaint website in just three days. The site wanted to “identify emotions” in their data, better understand consumer pain points, and improve services for the companies complained about on the site.

 
Describe the business challenge or problem you needed to solve

Visual Science has more than 10 years of experience in different process optimization and management solutions. They use BI and analytics in small, medium, and large companies in many segments. Focusing on live and visual management, they create custom studies based on customer data, turning the decision-making process into a whole experience.

 

A company that owns a popular complaint's website in Brazil, was concerned not only with closing cases but with understanding the consumer experience from end-to-end. The process takes a while, and some steps in between the case opening and the moment when a company responds (and take action, if necessary) needed to be understood. Several conversations are saved in text fields, and the sentiment analysis helps to identify how each consumer felt throughout this process. Additionally, it's possible to analyze other data, such as response time, how long the complaint lasted, days of the week, specific periods as Black Friday, age group, segment, device type, the volume of complaints, among others.

 

The website works as follows: The consumer registers and files a complaint against a company that has the right to defend themselves and a chance to solve the problem.

 
Describe your working solution

Visual Science developed a dashboard containing a variety of data, including text sentiment analysis performed using the Intelligence Suite algorithm to identify consumer subjective emotions and feelings.

 

  1. Authentication via access token + API

The first step was to connect the API generating a token. Next, access the API, insert token, and download the JSON to organize all the data. “If I needed to write code, the same process would take a lifetime, with Alteryx I did it in 10 minutes. I brought the URL to the workflow with the token, downloaded the JSON, and delimited points”, says Priscila.

 

API authentication via token and JSON download with informationAPI authentication via token and JSON download with informationAlteryx result returning tokenAlteryx result returning token

 

  1. Mining + Translation

The API brought a huge amount of data, more than 2,800 fields between metrics and dimensions. Mining and cleaning were necessary, especially in text fields. Each ID was separated by events, needing more in-depth work to concatenate based on the moment of each record. The sentiment analysis algorithm only understands English, so a simultaneous translation was necessary. We used a tool based on Bing Microsoft translator, available at Alteryx Gallery.

 

Azure translation appAzure translation app

  1. Sentiment Analysis

When opening a complaint, it's possible to write a free text explaining the problem. After the company's response, the person who opened the case can say whether it was solved or not. They also have the option to write a final message. Since the intention was to understand feelings from end-to-end, text was divided into different workflows for each case. Before running the algorithm, the text pre-processing tool was used to transform all fields into lowercase and to exclude the main stop words.

4.png

  • Consumer consideration in the input text

The sentiment analysis algorithm was used to measure the valence (whether positive or negative) and the magnitude (how positive or negative) for the emotion in the text, finding feelings in three levels: document, sentence, and characteristic.

This Tableau chart shows the opening text analysis over a period. We were able to verify the algorithm performance, which shows that consumers’ feelings are negative, however, improving with time. By clicking on the graphic, it’s possible to see a more detailed report.This Tableau chart shows the opening text analysis over a period. We were able to verify the algorithm performance, which shows that consumers’ feelings are negative, however, improving with time. By clicking on the graphic, it’s possible to see a more detailed report.

 

  • Consumer consideration in the output text

Using the same algorithm in a different workflow, the analysis shows the complaints' closing texts.

Here you can see the closing text analysis over a period. We were able to prove once again, the algorithm performance. It shows how consumers’ feelings became more positive at closing. We were also able to identify a negative peak, understanding cases when there was a negative feeling still.Here you can see the closing text analysis over a period. We were able to prove once again, the algorithm performance. It shows how consumers’ feelings became more positive at closing. We were also able to identify a negative peak, understanding cases when there was a negative feeling still.

 

The final graphic shows the outbound and inbound complaints analysis, with the average outbound feeling concerning the selected period. Exit texts are considered more important since they express the consumer's final feelings. Next to it, the entry graphic shows negative percentages as a bigger entry problem.         

 

7.png

 

  • Recognition of Top Words

A specific study was created to categorize models and identify the most used words, in addition to their relevance. The output shows the most relevant words in the entire database.

 

8.png

 

          4. Joining and Dashboard

This step joined all general information and feelings during opening and closings for each case. A Hyper output was connected to Tableau, continuing the dashboard development. The most interesting thing is that when you click on the feelings' graphic, the dashboard filters and brings only the information from that moment. This helps to segment the analysis by date.

 

Tableau dashboard with sentiment analysis resultsTableau dashboard with sentiment analysis results

 

Describe the benefits you have achieved

The website will use the sentiment analysis dashboard, offering a consultancy service to the companies mentioned in the complaints. Their goal is to identify the cause of the problems and improve their services. “Without Alteryx Intelligence Suite it wouldn’t be possible to make this complete sentiment analysis so fast. This solution allows managers to filter specific problems, being able to take action immediately”.

 

In the future, the goal is to also use text analysis on social media, since consumers are using it to express their feelings about a specific product or brand.

 
Related Resources
 
Comments
siddhartha0208
7 - Meteor

Thanks for sharing your fantastic use case and Alteryx work flow journey. Truly motivational.

crushgear
5 - Atom

Thanks for this wonderful story. It is really fantastic.

pujaguptars
6 - Meteoroid

Thank you so much for sharing.

Kaustubh17
8 - Asteroid

thanks for sharing wonderful case study it helps in solving real life problems.