Advantage

Saving time and costs through efficient questionnaire evaluation

Data

Open responses from more than 40,000 survey participants

Method

Gradient Boosting Machine


Automatic categorization of answers in customer surveys

Challenge

The OBI Group is the market leader in the German construction and DIY sector and, with more than 650 stores across Europe, is one of the best-known DIY stores. In order to continuously improve its range of products and customer service, OBI offers its customers the opportunity to share their feedback with the company by means of an online questionnaire. The open questions in particular provide valuable and often unexpected insights. However, in order to be able to evaluate the answers to these questions, they first have to be categorized manually before evaluation.

Goal

With more than 40,000 survey participants, manually assigning the answers to suitable, predefined categories such as product selection, location or price-performance ratio is very time-consuming. For this reason, OBI was looking for a solution to automatically categorize open responses.

Solution

In a first step, eoda prepared the answers submitted by OBI. In order to optimize the prediction quality, similar terms were condensed with the help of a word stem function, orthographically incorrect terms were assigned to a suitable term and meaningless words were excluded. In addition, relevant terms that frequently occur in combination with each other were combined under a meta-term and thus implemented in the model.

Following data preparation, a model was formed using the Gradient Boosting Machine data mining algorithm, which automatically assigns individual customer feedback to the appropriate response category. This model was applied to the open answers and the results were compared with OBI’s manual assignment.

Result

The model developed by eoda achieves an accurate classification of the answers. With its high accuracy, the model represents a valuable support for efficient questionnaire evaluation. Answers to open questions thus no longer have to be categorized manually, but can be automatically assigned to the appropriate answer classes. This leads to a significant reduction in the amount of work and time required while maintaining high quality.


We also realize your project