Ideas on analyzing sentiments


(Marine Maupin) #1

Hi everyone,

This is my first time posting on the forum but I have been reading a lot of it and trust you guys have plenty of good ideas. I’ve started thinking about anylising the sentiments of members rather than only looking at the amount of posts they create, I’m curious to find out what you guys are already doing and what you think! :slight_smile:


(Richard Millington) #2

Hi Marine,

Welcome and thanks for joining us. I’m a big fan of Robin and the insided gang.

Can you add a little more context here? What are you trying to uncover? What sentiments are you looking for? What’s the goal?


(christopher w) #3

Hi @Marine,

There are various cloud products available that can evaluate sentiment on social networks, blogs, websites, forums and so on.

They work by searching target urls for user entered keywords, example: VW Emissions, and reporting back on specific comments or general sentiment associated with that key word.

Governments, brands and PR agencies, amongst others, are keen users of this technology, with brands in particular reaching out to the most damaging individuals in an attempt to influence their public statements - free car service, product discount, vouchers and so on.

Microsoft have a very good product called Social CRM/Social Listening which is part of their cloud CRM offering - the following video gives you a good idea of its capabilities:

There are other options available too and searching the web for sentiment analysis will bring up a list of alternatives.

As an alternative to the standalone paid cloud products mentioned above, some sentiment analysis services are now being made available by application programming interfaces (APIs). You’d use the API by adding code to your forum which would send comments to the sentiment analysis service via the API. It would respond by sending you back an evaluation of sentiment - a simple number: 0 to 100, in the case of the demo below. Your code could then alert you if a threshold had been crossed - for example by emailing you a copy of any comment which has a score of 70 or less (where higher is better).

Here’s an example, again from Microsoft:

Machine Learning APIs: Text Analytics for Sentiment, Key Phrase Extraction, Language Detection and Topic Detection

Here’s the demo site - experiment by typing in sentences to see how it scores sentiment.

It’s early days for this approach but it looks solid enough to use in non critical environments.

Happy to advise further if the above is of any help.

Chris


Algorithms
(Marine Maupin) #4

@richard_millington Hi! great to hear you’re a fan of the inSided gang. I have been reading about social media and sentiments analysis and was wondering what where others already doing for communities (if anyone is measuring sentiments at all). The goal is really just get more insights on what can be done with it.

@ccdw Thanks a lot! Will definitely read more about the different products and play around with the demo you’ve shared.


(christopher w) #5

I’ve just noticed the API service is free for up to 10,000 transactions a month and correction, returns a score between 0 and 1.

https://datamarket.azure.com/dataset/amla/text-analytics