Analyze an existing NLP model to identify potential issues including training phrase conflicts, overlapping Intents, and duplicate phrases.
Experiment with new phrases to see how they compare to the existing model.
Cluster all training phrases across Intents to identify potential conflicts. For example, clusters of phrases that belong to different Intents, which may cause issues.
Cluster phrases within Intents to see the similarity of training phrases.
Identify phrases that are similar to clusters of phrases in a different Intent. These may cause Intent conflicts.
Overlapping phrases are training phrases that are similar to ones in other Intents. This may cause conflicts.
Identify training phrases that appear in more than one Intent.
Test additional phrases to see how they compare with existing training phrases and Intents.
When building an NLP model for an automated chatbot, a great place to start is with existing, live-agent log data.
Analyze existing chat logs into clusters based on semantic similarity. The clusters can be potential Intents and training phrases.
Build a model from the clusters and export for popular NLP Engines.
Analyze Existing Data
Analyze existing chat logs into clusters based on Semantic Similarity.
The clusters can help identify potential Intents and training phrases for an automated model.
Build and Export
Build a model from the cluster analysis.
Select and export the clusters as Intents and training phrases for an NLP Model.