Netskope provides a comprehensive data loss prevention (DLP) enforcement solution for cloud applications and public cloud resources that is ideal for addressing regulatory compliance requirements and protecting sensitive data in your enterprise.
In addition to securing traditional confidential and sensitive data on the web, Netskope DLP can also be diversified for some creative data analysis. In this blog post, I would like to share one of the use cases that can be implemented with Netskope DLP, which is sentiment analysis of user queries in Gen AI web apps.
Sentiment Analysis is a technique used to determine the overall sentiment expressed in a piece of text, such as a customer review, social media post, or email. Sentiment analysis relies heavily on positively and negatively used words to determine the overall sentiment of textual data.
To learn more about this topic, read the following and register for our upcoming webinar—where we’ll cover this live in more detail and answer customer questions. Let’s now look at how Netskope DLP can be used in conjunction with sentiment analysis.
Test Setup:
- Compile all the positive and negative words that are commonly used in sentences into two different CSV files. Many keyword datasets are already available on the web, which can be referred and used for this purpose. The CSV files would look something like this:
- Navigate to Policies > Profiles > DLP in the Netskope tenant console. Under DLP rules -> Entities, you can create two new dictionaries, one for positive words, and one for negative words and upload the csv files accordingly.
- Create two new DLP rules, one for positive words (include the dictionary that was created for positive words), and one for negative words (include the dictionary that was created for negative words).
- Create two new DLP profiles and include the DLP rules that were just created, one for positive words and one for negative words.
- Create Real Time policy to perform sentiment analysis on Gen AI category where users post questions and queries, and include the DLP profiles for positive and negative words, each with its own user notification, in case you would like to coach users. Other actions can also be selected as per the use case.
User experience:
We tested this out by trying to submit movie reviews on ChatGPT and this was the result:
Results and Recommendations:
As of now, sentiment analysis can be applied for user requests and queries on Gen AI and not the responses sent by Gen AI since there is no visibility for the response activity by the Netskope connector.
This can also be extended to other use cases like detecting social media trolling, abusive messages, criminal motive queries, providing affirmations to people who may be feeling low and so on.
We’d love to know your thoughts about using Netskope DLP for such scenarios.
