Infoland Sentiment Analysis

Project description

Research in what way, with the use of AI sentiment analysis, moderators can be helped to focus on messages that need attention because of a negative sentiment. Also consider that there may be more ways to classify messages than just negative or positive, this should be researched. It should be considered that Infoland cannot share a database that contain these messages. It is considered part of the project to solve this problem. You should also explore the possibility of making it a learning model by using feedback from moderators.

Thus, the main questions is: "In what way, with the use of AI sentiment analysis, can moderators be helped to focus on messages that need attention because of a negative sentiment?"

Context

Infoland is currently busy developing Zenya boost to sell this additional module to any customer using Zenya. Within this module they will implement a chat feature similar to Microsoft teams. This chat feature has to be moderated and the moderators / project leaders want to understand the sentiment of the project that they are spearheading. To achieve efficient chat moderation Infoland wants to use a sentiment analysis AI to determine what the sentiment from a message is and mark these to provide them to the moderators for an easy and quick summary.

Results

A small sample test set of 50 messages written in the style that should resemble the style of future messages sent within Zenya has been manually written by the client. This is of course not a very large sample, but should be enough to get a rough overview of the performance of each model. For more information about the performance of each model, it is stored within the model performance Excel. Below a small summary table is displayed.

About the project group

We are 4 students from Fontys, 2 from Business, 2 from Software and all 4 AI4 as a specialisation. We have worked on this project during this semester for 3 days a week.