Aspect-based sentiment analysis of video reviews

#hackwithdg #nlp #beginners

Introduction

I've always loved participating in hackathons, generally because they are a great way of showcasing my work, improving my skills and networking. I came across the Deepgram Hackathon on DEV a little bit late, that is a week after it was launched and with the approaching deadline, the Innovative Ideas category was the best option for me to be able to submit in time and not miss out. Another reason as to why I chose the Innovative Ideas category is because, I am still not advanced in programming and so it would have been quite hard for me to come up with code that would successfully complete the 'Build' challenge. Also, it's my first time participating in a Hackathon where I am to come up with an innovative idea. This is quite interesting and I'd really like to try it out.
This is my first time encountering Deepgram, though I do have basic knowledge of speech recognition technology and have encountered it in quite a number of occasions.

My Deepgram Use-Case

My idea is to create a model that will be able to perform aspect-based sentiment analysis of video reviews. The model will be able to extract the sentiment and classify the video review as either positive, negative or neutral. It will also be able to extract the aspect and classify the video review into the category, feature or topic that is being talked about in the video.
Customer reviews have been discovered to make a huge impact on whether a customer will make a purchase from a company or not, actually, much more than the marketing and advertising the company does on its brand. Having customer reviews is a great way for a company or a business to gain customers' trust on its products and or services, that is in the case of good and positive reviews. But in the case of negative reviews, then it is a good way for a company or business to understand the customers better and make improvements on their products or services. Customer reviews can also be used by investors for finance and stock monitoring. Investors can choose on a company to invest in by looking at the sentiments of the company's products.
Customer reviews have most of the time been written text, but of late video reviews are becoming more popular with customers. As it has always been said, 'People are more likely to believe what they see than what they hear.' Most video reviews are usually done on products, where someone gives his or her experience in using a product and gives a real-time demonstration of the product.
Because sentiment analysis is usually done on text, for this project, Deepgram will help with their speech-to-text technology by transcribing the speech from the videos into text after which aspect-based sentiment analysis will be done on the transcribed text.

Dive into Details

The main challenge that will be solved by Deepgram is to get the speech or audio from the videos into text format so that aspect-based sentiment analysis can be done on the text.
The main people who will benefit from this innovation are companies, businesses and investors. The main idea here is to automate the process of the sentiment analysis of video reviews. The benefits that the companies and investors will get from this innovation are:

Saving time. In the case of a large number of video reviews to be analyzed, manually analyzing the videos could take a lot of time. Something that could take a short time through automation.
Having a more trustworthy analysis. In most cases when performing analysis, humans tend to rely on their own experiences and unconscious biasness to derive meaning. The automated analysis will be able to remove human biasness through consistent analysis.
Having a more powerful analysis. They will be able to perform analysis without having limits on the data size to be analyzed.

This innovation will make use of Deepgram's model feature, with the option set to video, this is because the audios will be sourced from videos. Since this innovation is all about text analysis, Deepgram's keywords, utterances and utterance split will be very vital. With the keyword feature enabled, the model will be able to intensify a keyword or supress a keyword. Thus, enabling it to understand the context in the text. This a huge plus since the analysis will require a clear and a well understood context.

What lead me to this particular idea is that while searching for inspirations from Deepgram support page, I came across this project by Kevin Lewis, where you would have a wearable screen that live captions your voice to help people understand you while wearing a mask. Suddenly, it rang in my mind that there's actually a lot we can get from reading texts, the sentiments being relayed, topic and context being discussed. And that is how I ended up coming up with the idea of aspect-based sentiment analysis of video reviews.

Conclusion

By participating in this challenge, I have really got to learn a lot about Deepgram, its features and how they can be used. It has also been really nice to go through the other participants' posts and learn so much about the several ways in which speech to text technology can be used.