About Genius Text Analytics

The Genius™ application is a Forsta add-on and subject to licensing fees.

Important
This functionality is only available for SaaS customers.

Forsta Genius Text Analytics is a text analytics application that analyzes text responses to survey questions, groups them according to subject, and grades them on a 5-point scale (-2 to 0 to +2) according to how positive or negative the respondent is to the subject of the question.

For example, a shop could ask customers what they thought of the service they received while making their most recent purchase, and a respondent's comment could read: "Your employees were very knowledgeable and helpful, but the floor was dirty and the shop was rather untidy." After processing by Genius, the result could be that Staff knowledge is graded at +2 and Staff helpfulness at +1, while Cleanliness is graded at -2 and Tidiness at -1.

Your survey can contain as many questions of as many different types as you wish (within system limits). However text analysis can only be performed on open text questions.

The text analysis is based on a categorization model. Different techniques are available for helping you to build your categorization model, using advanced Boolean expressions, Concept Miner tags or a combination of both. This model must be created, and must contain all the key-words and tags that are to be searched for in the responses and graded during the analysis.

To create your categorization model you need access to Model Builder, which is our self-service tool for understanding your verbatim and for creating the model. Concept Miner is a topic extraction tool that uses AI to analyze your data and extract the relevant topics. These topics can then be used to create tags, which can be brought into your categorization model for quick and easy model building.

Model Builder is subject to per user licensing fees and has separate documentation.

How it Works

The Forsta Genius Text Analytics add-on allows categorization and sentiment analysis of free-form text (open text responses).

In order to set up Text Analytics, you will need to have a categorization model that has been built for your business and which is specific to your verbatim set. Depending on your situation, you would fall into one of the following groups:

• You already have a categorization model from a different vendor – we are able to import your model into Model Builder (our end user model categorization management tool).
• You have the outline of a categorization model from manual categorization or coding work.
• You do not have a model, but your business falls into a generic model vertical that Authoring has already created.

Once you have your categorization model, the Forsta Genius Text Analytics folder must be added into your survey, linked to the model and sentiment scale and the Authoring task scheduler set up to run at regular intervals. Forsta Genius Text Analytics is tightly integrated with Authoring, so that any new verbatims that come in through the survey responses are sent across to Genius for categorization and sentiment. These are stored within the Genius folders so that the reporting modules can use them for reporting and dashboards.

Note: If you want to avoid personal data / personal identifiable information being included in the Verbatim, it is recommended that you add an instruction in the open text question(s) along the lines of the following: “When providing your response, please do not include any personal identifiable information".

Authoring Integration

Once the model has been built, your survey responses can be analyzed.

When you run Genius Text Analytics (go to Running the Genius Task for more information), a task submits verbatim texts from the Forsta survey to Genius. The texts are processed through the Concept Miner categorizer and the sentiment agent, and the analyzed data is returned and stored in the survey database. Overall sentiment, category hits and category sentiments are stored into regular variables in the survey, making them available for reporting and analysis. These variables must be set up in the survey prior to the analysis, for each verbatim question for which the answers are to be processed .

Once the responses have been analyzed and you wish to present the data in your report, bear in mind that text analytics deals with multidimensional data; categories mentioned and their individual sentiments. The challenge here is to present the multidimensional data in a way that:

• Is easy to comprehend at a glance.
• Draws attention to the areas that need changing, where many people have negative experiences.
• Avoids real issues becoming hidden behind averages.

For example, say we have a survey with 1000 responses. 400 of the responses talk about employee attitude, with only 25% (100) of those statements being negative. This then makes the Average sentiment for attitude very positive. If 50 of the responses talk about store layout with 80% (40) of those statements being negative, then the average sentiment for layout is very negative. If we then report based on average sentiment, we will fail to notice we have 100 customers (10%) that have negative experiences with employee attitude, and rather focus on improving store layout because of the 40 customers (4%) that have negative experiences with that.

Also, overall sentiment has a tendency to end up as neutral because people often mention both positive and negative things in the same sentence, for example, "The service was good but the shop was untidy". This is where category sentiment becomes important.

It is also important to think about how you are asking your question. The survey question should be neutral, such as “Please would you tell us more” or “Please let us know your thoughts about your experience”. This way, respondents will write sentences that contain sentiment that can be analyzed.

If you ask a question that is already loaded with sentiment (for example “What did you like about your experience in the store?”, the sentiment engine will not be able to provide accurate sentiment as the respondent is likely to reply with a neutral verbatim.

These points must be considered when creating the reports.