Adjust Text Analytics (Sentiment) Model

asked 2021-05-09T14:22:45.267+00:00
Justin 26 Reputation points

I am new to Azure services and just testing it out at the moment.

We are using cognitive services, specifically text analytics / sentiment detection to try and determine the sentiment of short phrases.

I am curious if there is some way to adjust the model being used for smaller amounts of text. Most of the text we are analyzing are about 140 characters long (short).

It seems like the sentiment analysis is tailored towards longer form text. Are there any options to set it to "short text" or to help train the model when we see "incorrect" analysis. I am just trying to find out if we can fine tune the service in any way.

Thanks in advance!

Azure Cognitive Services
Azure Cognitive Services
A group of Azure artificial intelligence services and cognitive APIs that help build intelligent apps.
936 questions
Cognitive Service for Language
Cognitive Service for Language
An Azure service that provides natural language capabilities including sentiment analysis, entity extraction, and automated question answering.
173 questions
No comments
{count} votes

Accepted answer
  1. answered 2021-05-11T11:24:07.8+00:00
    Ramr-msft 10,061 Reputation points Microsoft Employee

    @Justin Thanks for the question. Can you please add more details about the Text analytics version that you are trying. The models are pretrained. We do not host custom models. If you want to create and host custom machine learning models, consider the machine learning capabilities. Please follow the Text Analytics API (v3.1-preview.3) API returns a detailed sentiment analysis for the input text. The analysis is done in multiple levels of granularity, start from the a document level, down to sentence and key terms (aspects) and opinions.

    If you have a real dataset (preferably labeled but unlabeled is okay too if you can review the results with us) in any of the supported languages Language support - Text Analytics API - Azure Cognitive Services | Microsoft Learn and have interest in trying out our new Transformer based pre-trained AI models (A holistic representation toward integrative AI - Microsoft Research) – can you please reach out to me to evaluate it?
    Your dataset will not be used to re-train the AI models, but we can fine tune the model if the results are not as desired.


1 additional answer

Sort by: Most helpful
  1. answered 2022-03-25T16:25:23.19+00:00
    KR 1 Reputation point

    Hello @Ramr-msft - I have a use case for the fine tuning opportunity you mentioned. Should I start a separate thread to start the evaluation process? Thank you!

    No comments