Within an Email Support an environment topics and resources are often stored within knowledge articles. These articles are typically created on an as-needed basis and then attached as emails come through. Due to way Knowledge Articles are created, there is a heavy importance on agents reporting and suggesting new articles. However, for agents there is a constant trade-off between doing cases and reporting what's needed. Consequently, we have a number of large KAs that have a high case count but are considered a 'catch-all'. From a reporting perspective, these KAs likely dominate reports despite haing little value.
This repo focuses on using a Clustering Algorithm on these large KAs to see if there are potential segments within the KAs that we can split into new KAs for better reporting and for faster resoultion times of cases by having a dedicated KA and template. Depending on the volumes of the segmentted KA, there may be the possibililty of automating responses to it as well.
Clustering is a machine learning technique that groups objects into clusters based of how similar they are compared to other clusters. It's an unsupervised learning technique which means we don't need to input labelled data such as we've done in the Auto-answer analysis - this saves money in terms of labour required to label data.
- cluster-example.ipynb - Clustering using example data from Sklearn's dataset package
- feature_request_clustering.ipynb - Clustering using actual case data to identify topics of cases sent that were assigned to a generic KA
- presentation_code.ipynb - Jupyter notebook presentation based on the feature_request_clustering.ipynb notebook. Steps on how to produce presentation are included in the notebook.