Deep Sentiment Analysis of Product/Movie Reviews
This application uses Convolutional Neural Nets (CNNs) to do real time sentiment analysis as you are writing the review. For detailed sentiment analysis, IBM watson Tone Analyser API is used. Real time analysis happens on the browser itself, is privacy preserving and does not make any network request. Detailed Analysis uses Support Vector Machines (SVMs) and makes a network request to watson Tone API.
Often on social media, e-commerce websites people (especially non-native english speakers) have a hard time conveying their message/comment/review in their intended tone/meaning. This application does a thorough sentiment analysis on a piece of text to give more information to the writer/user about the underlying tone of their writing/post/review.
This idea can be used in:
1) Review Monitoring/Enhance customer service (if somebody posts a bad review, you can get notified and take some remedial action)
2) Social Media Monitoring/Social Listening (if somebody is posting a bullying message, you can warn them)
3) Integrate with Chatbots/AI (chatbot can sense the underlying tone of the conversation and if the user is unhappy, take some remedial action)
4) Group chats (To maintain/monitor a decent tone in the conversation)
5) Official mail and communication (To check whether the e-mail written by user has proper/decent tone or not)
Note that the training is done on IMDB review(s). So the longer your review is, better the analysis will be. Analysis for short sentences might not be that good.
Write/Copy a movie/product review in the text box to get real time sentiment analysis. Press Detail Analysis button to get more advanced Sentiment Analysis.
View the screencast:
Node.js, Angular, D3, Tensorflow JS, IBM Watson Tone Analyzer API
Pretty neat. I'd love to get started with machine learning. I'm sure stuff like this will be very useful across a broad range of sites and apps.
Hm, interesting idea, but the execution is lacking something.
For one, I was confused, why the explanation text moved up and down. Why not put it under the buttons below the textarea?
The emoji went left and right … Does it correlate in some way with the charts on the right hand?
What I'd expect is some explanation what is happening.
For example, which amount of text was used for analysis (you could give it a background colour).
I was originally planning to work on something similar but got an even better idea (in my opinion) shortly before the hackathon started.
Nice work. Would like to learn more about the implementation. Could you write a blog post or something?
It's a great concept and executed nicely. I liked the detailed analysis and wished that was shown right off the bat.
It is kind of easy to fool if you're trying though!
This is pulling together some pretty cool APIs and services. I think a little more work could have been done to help the user in the intended use case. For example, if I really wanted to use this service to help shape my tone in a non-native language, it might be nice to know what parts of my text contribute most to the overall tone.
Nice work given the short time and modern, complex tech stack involved. I like the idea and it seemed to analyze my reviews pretty well. For some reasons I couldn’t get the “real time” emoji to really work (maybe my sentences were in fact too short!?) but all detailed analytics were spot on. Solid base work and I’d be curious to see this idea being taken further. Sticking to reviews the analysis could help to identify the general tone, vibe and common “feelings” of reviews (movies, music, restaurants) and accompany the usual “star rating”