Releasing a Minimum Viable Product (MVP) is nerve-wracking. Our early stage product features a platform where you can use a sentiment analysis AI with any text-based data you have (think: blogs, emails, texts, stories, social media feeds). We will also allow you to build a sentiment analysis AI application without having to code and using any text-based data. Is our platform beautiful? Don’t hurt its feelings! Is our platform full of bugs? Perhaps. Are there embarrassing spelling errors? I hope not but maybe. Our closed beta is far from perfect or even far from what the final product will be but we’re so excited to have announced and started emailing our sign ups their invitations.
“If you’re not embarrassed by your MVP, then you launched too late.”
- Our CTO's favourite quote from someone else
How we decided which AI task to add first on Metaranx
We had to make the difficult decision of which AI task we would release first under Metaranx. Releasing as many AI tasks as possible would not be a good choice: over-building anything in the early stages of a startup is never recommended. We settled on starting with language-based tasks and our first will be sentiment analysis. This decision was based on what our users stated they were looking to use AI for, what we thought would cost us the least to operate (text-based datasets are less expensive than image, video, and audio-file based ones), and what had a broad range of applications for users.
Sentiment analysis can accomplish a lot of what you need - it’s much more capable than a simple “negative” or “positive” labeling of text. Sentiment analysis, as with many AI tasks, is influenced by your dataset and how you structure that data. When you build a sentiment analysis AI application, your data and how you structure that will influence your application a lot.
What is sentiment analysis?
Sentiment analysis is also referred to as opinion mining or emotion AI. Sentiment analysis uses Natural language processing (NLP) methods and algorithms. NLP refers to how we program computers to process and analyze large amounts of natural language data. NLP can help understand the structure and meaning of text and you come across them daily in email filtering, chatbots, predictive text, search suggestions, and more. The artificial intelligence component of this is training the computer to make decisions based on the natural language data. Sentiment analysis uses natural language processing to identify, extract, quantify, and study affective states and subjective information. With sentiment analysis, you can determine patterns in any body of text or dataset of text: whether you want to focus on mood, emotion, attitude, and more.
How to use the sentiment analysis AI application on Metaranx
The sentiment analysis AI available on Metaranx will tell you how "negative" or "positive" your text is. All you need to do is upload your dataset to Metaranx via a CSV or copy and paste text into your console. From there, click "Get Sentiment" and all your data will have a positive or negative sentiment attached to it. You can download this as a new CSV or store it in your console. If you don't have data but want to play around, pull a free dataset from the marketplace and use it with the sentiment analysis AI to see how this task works. You just used AI and hopefully it helped with your business or personal projects!
How to build a sentiment analysis AI application on Metaranx
Sentiment analysis is not restricted to measuring the positive to negative sentiment of statements. It can help you measure between any categories of any text-based data - it’s all about how you structure your application and your dataset.
- Choose the sentiment analysis task
- Upload your dataset or choose a dataset from the marketplace and add it to your Console’s blank Canvas
- Time to structure your data! Decide what labels you want to create that aligns with your text.
- Train your sentiment analysis AI. Training time and costs vary on the size of a dataset (max 10MB).
- You’ll receive a notification that your AI has finished training. Test it out!
- Publish to the marketplace for sale or open source and use it right away.
In many sentiment analysis datasets, the labels are “positive” and “negative” but maybe you want to build an AI that can determine if a social post will be “successful” or “not” or have a few different labels to see if your customers are responding as: “happy,” “neutral” or “angry.” Maybe you’re labeling your job search data as “job,” “volunteer,” and “internship.” The more labels you have, the more data you need to teach the AI what those labels mean. Set your labels to the corresponding text so that your AI can learn what you’re labeling and why.
The best way to think about structuring your data for sentiment analysis is: how would you teach a child what this sentence, paragraph, or other content means within your labels? You would show them many different examples of “sad” versus “happy” content so that they could learn. The more examples they have, the better chance they have of getting it right when they are presented with new information and have to define it as “happy” or “sad.”
Examples of sentiment analysis AI applications you can build
Sentiment analysis for customer responses to your business or brand
You handle all customer-facing initiatives for your business. You want to determine the sentiment of all new incoming messages, reviews, posts, and support queries are “positive” or “negative.” You would train (specifically, finetune) a sentiment analysis AI on your company’s data. This dataset of your company’s existing customer messages would be labeled as “positive” or “negative.” After training, when you enter a new sentence, the AI should give you an accurate reading on whether it’s positive or negative. You can make better business decisions on products, announcements, customer support scripts, social posts, and more with this information.
Sentiment analysis for better social posts
You run all social media channels where you work. You see some posts are more successful than others, but can’t seem to figure out why. You upload all your Tweets and label them as “successful” or “not” based on their number of likes. You would train (specifically, finetune) a sentiment analysis AI on your workplace’s labeled Tweet data. Now, when you enter a potential Tweet into your AI application, it should know whether or not that will be a successful Tweet or not before you publish.
Sentiment analysis for sifting through job posts
You’re looking for your first job and you’re open to anything that will give you solid experience in a field you enjoy. Maybe you’re labeling your job search data as “paid” and “internship” so that you can make decisions about what you want to apply for. Maybe you’re looking for work at a law office and don’t want to see anything else so you choose labels like “law” and “other.” Use a dataset with job postings and label them according to your chosen labels. Train (specifically, finetune) the AI with your dataset. Use your new AI application to run through new lists of job postings to filter the ones you want to apply for.
Sentiment analysis for finding fake news
It's difficult to know which news is not accurate, spreading misinformation, or written entirely by AIs with facts that aren't true. Get a fake news dataset labeled with news that is "real" and "fake." Train (specifically, finetune) the AI with your dataset. Every time you are unsure about news you're reading, put it into this AI application and see if the AI believes it's fake or real news.
How you can help us grow Metaranx
We’re truly grateful to all our users and the enthusiastic community that surrounds no code, artificial intelligence, data science, and entrepreneurship. When you’re trying out the Metaranx platform, we would love it if you could send a message through our Live Chat feature that exists on the website (bottom right!) and in the console or through social media, contact form, my email - go wild! We’re looking for any of the following and welcome all comments:
- Feedback pertaining to the use of the platform
- Was it easy or difficult to use the AI?
- Was it easy or difficult to build an AI?
- What would make using an AI and/or building an AI easier?
- Feedback pertaining to the marketplace?
- Was it easy or difficult to find the data set(s) you want?
- Where did you get stuck on your path to using or building an AI?
- What AI do you want to build next or what do you want AI to accomplish for you?
This feedback helps us add new features and determine which AI task we will add next. We got feedback on our release day from a user saying “What data can I put in the sentiment analysis AI?” - good question! So, on July 2nd Lisen got to work on the Marketplace. There will be a few open source (free!) datasets that can be used with sentiment analysis if you don’t have your own but want to test the product out. If you’re looking to test it on your own content, try copying-and-pasting or making a CSV file (through Excel or Google Sheets) of your:
- Blog posts
- Content titles
- Social media posts
- Text messages
- Stories or creative writing
- Any text-based item that you can copy-and-paste or turn into a CSV with ease
How I would personally want to use sentiment analysis
I tend to write in the “negative” a lot. It’s not something that I intend to do (example: this sentence could be reworded as, "I strive to write positively") but I often don’t realize that my emails, social, and content comes across that way. Sentiment analysis could help me understand how negative my blog is perceived, so that I could remedy it if necessary prior to publishing. Using future Metaranx AI tasks, such as entity extraction and detection, I could locate those pesky negative sentences and rewrite them - maybe with a content generation AI task! Hmm… that’d make a great combination for an AI application on the Metaranx platform.
This was insight into why we chose sentiment analysis as our first AI task for our closed beta release. We also discussed how you can use sentiment analysis for your business or what type of sentiment analysis AI applications you can build without code on Metaranx. The bottom line is that it’s all about your dataset and we’ve made it easy for you to use datasets or add your own datasets and make them ready for your AI on our no code platform. We want to know what you build, how you built it, and how we can keep making Metaranx better so you can achieve your AI goals. We can't wait to see how you use sentiment analysis but also to hear about what AI tasks you want to build next. Sign up for our beta - invites are slowly going out every day!