Reporting is easy when you have statistics. Numbers, frequencies, and graphs are the bread and butter of quantitative analysis. What happens when you have qualitative data that needs processing though?
Open-ended, free form responses for any survey can provide great insights. Comments offer explanations and paint pictures that numbers and graphs can’t always provide. But having over a hundred, maybe a thousand text responses to go through and complete a content analysis on is not an easy task.
SoGoSurvey has several reports that work with open-ended responses:
- The Individual Report display a single participant’s complete responses, including any text fields
- A Verbatim Report will display any data collected in open-ended questions in both card and word cloud format.
- The Response Table Report allows you to view all your data, including open-ended responses, in a tabular format in groups of 100. (Export the full set in Data!)
- The Omni Report allows you to see a visual representation of your open-ended responses with a word cloud alongside the graphs for your numerical data.
Each of these reports offers great opportunities to view a comprehensive list of all your open-ended responses. The Verbatim and Omni report even provides you a powerful visualization of how frequently certain words appear in responses. The bigger the word in the cloud, the more often it was mentioned by participants.
Reporting With NLP
The one thing that none of these reports can do, however, is analyze all those text answers. You may be able to see what everyone said, but how do you quantify qualifying data? That’s the beauty of SoGoSurvey’s newest report: Text Analysis.
Using Google’s impressive Natural Language Processing (NLP) capabilities, the Text Analysis report will, according to Google, help you recognize insights from unstructured text using Google machine learning. NLP is defined as “a branch of artificial intelligence that helps computers understand, interpret and manipulate human language.” It’s a way to bridge the gap between binary languages and the human ones.
When running a Text Analysis report, the platform will take all your free form responses and run them through Google’s NLP. Most reports can generate results in minutes. A few clicks through the wizard and you have a great report at your fingertips. Text Analysis, however, takes a bit more time to generate because of the AI training that must be done. Our users survey a variety of subject matter and fields—from healthcare to hospitality. To make sure that your content is categorized by relevant topics or tags, the SoGoSurvey team works on the backend to train the AI, making sure that each topic is relevant to the survey’s subject.
The first thing you’ll see in a TextAnalysis report is an Overall Metrics card at the top of the report. This is essentially a detailed summary of your open-ended responses, including metrics like:
- Total Responses
- Total Records
- Positive Responses
- Negative Responses
- Neutral Responses
The first tab of the report is the Dashboard, which shows a graph and a word cloud. The word cloud may look familiar to those who’ve seen it in Omni or Verbatim before. By clicking the three-dot icon to the top right of it, you can customize and filter your data as needed. With the ability to filter the word cloud, it provides a great overview of some of the more frequent responses your questions received. But how do you know if those responses are good or bad?
The key functionality that sets this report apart is its ability to assign a sentiment score. Depending on the NLP analysis, an answer can be scored as a negative, neutral, or positive sentiment. A sentiment score below -0.05 is negative and above 0.05 positive, any number between these is considered neutral. A maximum of seven topics carefully developed to be survey-specific are then assigned to each response. More than one topic can be assigned to a response. These same categories are found in the sentiment graph on the Dashboard.
The value of sentiment scores takes the guesswork out of all your free form questions.
The sentiment graph, a brand-new update that came out with the 19.0 release, is found to the left of the word cloud. This graph provides a visual representation of the overall sentiment percentages for each topic in easy-to-read donut charts. The graph displays a donut for each category or topic the responses have been divided into. You can see a complete list of responses for any chosen topic by clicking on the donut chart or by going into the second, Open-Ended Responses tab on the report. To pull out the details of the donut charts, just hover over the different colored sections to dive deeper into:
- Number of responses categorized within this topic
- Mean Sentiment
- Median Sentiment
- Number and percentage of positive responses
- Number and percentage of negative responses
- Number and percentage of neutral responses
On the Open-Ended Responses tab, you have a variety of ways to drill down on responses and find specific insights. Here you have the option to filter responses. You can filter by different categories, question numbers, topics or sentiments, all by clicking the column title and choosing from the dropdown menu. Doing a workplace survey and want to filter by department? Done! You can also search for more specific keywords in the search box at the top left of the chart.
What’s the one thing this amazing report can’t do? Like some humans, it can’t detect sarcasm. So, if you have a few sarcastic comments in the bunch, the system won’t be able to tell—yet! But coming soon there just might be the possibility to override the system’s lack of humor.
Whenever you’re ready, you can export this report for your viewing pleasure later. So next time you have a thousand comments with hidden gems of useful information, save time and use the built-in sentiment analysis capability of a Text Analysis report!
If you’re interested in learning how to add the Text Analysis Report to your account, or want to learn more about it, you can request a demo.