Dust Community Icon

Improving Data Retrieval for Intercom Conversation Sentiment Analysis

·
·

Hello here, I am working on a agent that analyzes Intercom conversations (which is our customer support platform) through the Intercom integration. My goal is to perform sentiment analysis on issues perception from clients to evaluate urgency, so we can enhance how we prioritize and schedule fixes. The results are promising and accurate but I have an issue of limited retrieval of data sources preventing me from conducting thorough analysis (cf. screenshots). I tried several models (gemini, mistral, gpt*, etc.) but got the same behaviour. Is there a workaround to enable fetching more data to the context ? Am I doing something wrong ? Tips or guidelines are welcome Thanks! Note: we are on the Dust Pro tier.

  • Avatar of Sébastien Hoareau
    Sébastien Hoareau
    ·
    ·

    I'm doing something similar in Zendesk, though maybe using a different approach:

    • I run a sentiment analysis on each ticket when they are closed.

    • I store the score and a small comment on the ticket itself.

    • I then use Zendesk reporting capability and alert systems for day to day usage.

    That way I never have to retrieve a large number of document.

  • Avatar of Remi
    Remi
    ·
    ·

    Hey there 👋 Thanks for your question! In this case you probably want to use the extract data tool instead. This guide can help: https://dust-tt.notion.site/Channel-highlights-a43da570bd154a7d89d42c7e738b9c5a?pvs=74

  • Avatar of Remi
    Remi
    ·
    ·

    I run a sentiment analysis on each ticket when they are closed.

    Sébastien Hoareau do you use zapier & dust to do it?

  • Avatar of Sébastien Hoareau
    Sébastien Hoareau
    ·
    ·

    n8n but yes

  • Avatar of Khaled K
    Khaled K
    ·
    ·

    Remi thank you for the documentation and the tips ! I was using the include data And I switched to search which is already giving much better results ! i'll give extract data a try when i'll start uploading data sources with transparent contents to reduce the schema.

  • Avatar of Khaled K
    Khaled K
    ·
    ·

    Sébastien Hoareau that is a clever approach. Thanks for sharing