Overview
Knowledge demand insights uses machine learning to automate the discovery of knowledge gaps in your knowledge bases. Knowledge gaps are identified when incoming cases are not matched to an existing active knowledge article.
Knowledge demand insights has the following benefits: (1) improves the effectiveness of
self-service by ensuring knowledge articles exist that can address current cases (2) Speeds case resolution by having relevant knowledge articles to deflect future cases (3) improves case resolution by recommending knowledge articles that can best address the current worked case.
The Knowledge Demand insights process first runs similarity and clustering to combine similar cases then compares those cases against existing knowledge articles. If the process doesn’t find a knowledge article that can solve the similar cases, it generates a Pareto chart via a scheduled job.
Note – If you enabled the auto training from article 1 some of your PI models may be already trained. You know the model is trained when you see a [version #] next to it. If that’s the case, you don’t need to re-train just move on to the next step.
Lab 4 – Configure & Train Knowledge Demand Insights
Train the Similarity Algorithm
1. All > Predictive Intelligence Homepage > Similarity
2. Expand the Similarity caret and locate the “Similar Knowledge Articles for HR Task” solution. Click Train. Depending on the amount of data in your instance your model may take 15-30 minutes to complete training.
3. Confirm that the Similar Knowledge Articles for HR Task has completed training, this may take up to 30minutes. The waiting in training status should be replaced by a [Version : #] when training is completed.
Configure and Execute Knowledge Demand Insights Job
4. Once similarity is trained we can kick off the knowledge demand insights job. This will generate the Pareto chart used to identify gaps in knowledge.
5. Go to All > Knowledge Demand Insights >Demand Insights Configuration
6. Click on sn_hr_core_case to open the configuration. Notice that the Demand Insights application uses a similarity and clustering solution definition.
7. Modify the Query condition as necessary and hit update. In my case the filter conditions are fine.
8. Go to All > Knowledge Demand Insights >Scheduled Jobs
9. Click the [Knowledge Curation]: Generate HR Case Clusters
10. Click Execute Now in the Scheduled Script Execution Form. This will execute the job that combines the similarity and clustering results into a pareto chart. The job may take up to 30minutes to complete.
Using the Knowledge Demand Insights Pareto chart
The Knowledge Demand insights application generates a Pareto chart identifying patterns of potential knowledge gaps.
1. Got to All > Knowledge Demand Insights > Demand Insights for HR Cases
2. If there is valid data, the Pareto chart will populate like so.
3. Each bar represents a cluster of similar cases which do not have knowledge articles.
4. Click on the first bar in the chart to show the case details.
5. Notice at the top you can Generate a representative sample or ignore the collection. Ignoring means you don’t agree with the prediction and you wish to hide this cluster.
6. Click the “Generate a Representative Sample” then check the select all check box to select all the cases.
7. Then click “Report Knowledge Gap”. This will create a knowledge task with the gap description for the knowledge manager to pick up and review.
8. You should see Feedback Task has been created.
9. Click on the Feedback Task Link. Demand insights has created a Knowledge Feedback task as part of the knowledge workflow.
Congrats you’ve used machine learning to identify gaps in your knowledge base.
Trouble Shooting - Lab 4
If your Knowledge Demand Insights Pareto chart comes up empty you may need to define a new similarity solution definition to fit your data.
1. Configure a new similarity model and train it.
2. Go to All > Predictive Intelligence > Similarity > Solution Definitions > New. Create a new similarity model and train it. In my example below I reduced the inputs and removed all the filter criteria.
3. To use the new similarity model you need to swap it with the OOTB similarity model used in the knowledge demand insights.
4. All > Knowledge Demand Insights > Demand Insights Configuration
5. Select sn_hr_core_case
6. Change the similarity model with the one you trained from step 1, in my case "Lener Demand Insights HR"
7. Kick off the Knowledge Demand Insights training job by going to all > knowledge demand insights > scheduled jobs .
8. Click into Demand Insights: HR Case Clusters Need Knowledge and EXECUTE job.
9. Click into [Knowledge Curation]: Generate HR Case Clusters and EXECUTE job.
10. Go back to All > Knowledge Demand Insights > Demand Insights for HR Cases to see if the Pareto chart is populated. If not you may need to open a support case for further assistance.
Note, all of the predictive intelligence solutions have minimum settings. I can't stress that you won't get reliable results if you test under 30k cases. However, I realize there are situations where you may just need to prototype a PI model and you get the below type of error. If necessary you can reduce the minimums needed for Knowledge Demand insights by going into
1. Login as admin and go to All > sys_properties.list
2. Go glide.platform_ml.api.min_similarity_window_records and reduce the default.
3. Go to glide.platform_ml.api.min_clustering_records and reduce the default
4. Retrain
