What are effective techniques for sentiment analysis of customer feedback for product design using NLP?

Based on the purpose of your product or part of product that you are designing/ improving you should be able to differentiate the kind of sentiment analysis that can yield best results. Should it be emotion detection, intent based or aspect based analysis that you should focus on. And then more or less the steps for analysis should be generic starting with data collection and removing noise. Followed by extracting features and selecting a machine learning model, rule based, automatic or hybrid one, there are lot of COTS tools available in the market to support this analysis. And then it's upto you, based on the context of the problem and how you want to view the results you can classify the analysis.

Personally keyword extraction still remains the best option for sentiment analysis. Feature extraction using LLMs has been a trend I have been observing around this space. Embeddings for known sentiments/set of words will definitely be the go-to process for sentiment analysis in near future.

It's also worth mentioning that it is quick and easy to use tools like Pendo to focus on a Net Promoter Score (NPS). 

First of all it depends on your use case as there are various ways by which you go about doing this sentimental analysis - of which most common one is having scaling based on emojis which depicts your behaviour/ emotion towards that particular query asked.

If your survey form corresponds to say a claim process submission feedback then there this emoji based scaling may work if customer did receive proper claim amount and process was smooth and nothing happened to patient BUT may not work in scenario where patient died even if process flow was smooth

So, first you need to be specific about your use case and then decide your target audience and correspondingly create analysis form

Next main and foremost important thing is to have a plan B like suppose if you launched a new product and want to get feedback on priority irrespective of result so that you can improve it, then you should know your target audience to be chased and what kind of audience this will be.

You can identify the nature of user basis his past records and then send corresponding form 

Using machine learning and artificial intelligence to segregation of positive, negative and neutral keywords from the feedback and with some suggestions