In this article, you will learn how to summarize and generate themes of conversations at any data point using AI+ Summarizer. 

​

Drawing insights by drilling-down at any data point can be a tedious task, as there might be a lot of mentions to go through. Sprinklr AI+ powered Summarizer empowers users with dynamic insights as they drill down the dashboard, revealing key pillars of conversation themes, action-oriented points, and concise summaries.  

Navigating AI+ Summarizer 

To summarize a data point while drilling down –  

1. At any point in the dashboard, while drilling down, click the Generate Summary button and a summary will be generated automatically.  

2. Detailed summary can be generated by clicking on the View Details button.  

3. User can provide summary feedback by giving thumbs up/down 

Generate Summary Button:

Popover Summary View:

3rd Pane Summary View 

Components of AI+ Summarizer 

The AI+ Summarizer consist of three view – popover, 3rd pane view & Standard Drilldown. 

1. Popover View – This view provides a glimpse of the conversation stream & displays summary points with a title, short description & an associated sentiment. 

2. 3rd Pane View – This view provides a detailed summary of the conversation stream and displays summary points with a title, long description, associated sentiment & the top entities. 

3. Standard Drilldown – AI+ Summarizer is integrated with the Standard Drilldown & provides a “detailed” summary of the conversation stream. 

How it works? 

1. Stream of mentions from different sources are used as an input. 

2. Sorting is based on source wise relevant metric; hence only the trending/relevant mentions are sampled.  

3. Source distribution in input messages is logarithmically sampled to eliminate source bias. 

4. GPT-3.5-turbo-1160 Model is being used, which considers 16k tokens, including the prompt & response.  

5. Messages are feed to GPT-powered LLM model to generate summary. 

6. Summary is generated. 

Current technical nuances & limitations  

1. It might take some time for the summary to load. 

2. We're using the GPT-3.5-turbo-1160 Model, which considers 16k tokens, including the prompt & response. So, only content within this limit is processed in a single API hit. 

3. The entire summary is generated by GPT, making it susceptible to differences at the same data point due to the model's tendencies for hallucination, ambiguity, and interpretation. 

4. The Drilldown Summarizer isn't compatible with non-hyperspace widgets like Smart Theme Explorer (Smart Clusters), Tree Map, and widgets displaying single mentions like stream and inline editable table.