HACKER Q&A
📣 heshiming

Open-source sentiment analysis toolkit, a valid small business idea?


I worked for years as a freelance contractor in NLP. Clients and agencies generally consider sentiment analysis a simple matter. But the reality is off-the-shelf packages, like TextBlob are rule-based. When you put together an app, like a chatbot for customer services, they rarely work well. The cost to improve is nothing short of reinventing a new system, which most wouldn't accept. Plus, the consensus is that people prefer self-hosted in-house solutions they can improve. Reasons to avoid sending texts to an external API are 1, privacy, and 2, per-call pricing.

So I have this idea. I create an open-source toolkit for sentiment analysis. It will be data-driven, not rule-based. So it can fit most use cases well. It'll be a full package with GUI, for labeling, training, and an API server to self-host the model.

I keep a private repo of data to generate a good-enough model for a niche: customer service chatbots. A part of which is curated (scraped), and a part is labeled by myself. I sell a subscription to the model, during which period, I try to gather and label more data to improve the model. I can design the system to support partial training. That is, the customer can improve based on my model, using just hundreds of lines of their own data.

I figured if I priced it the right way (10-20% of hiring someone to create one), I can sell it to dev shops and agencies. I can go to Linkedin or Upwork to approach my customers.

Does this sound like a viable idea for a one-man shop?

  👤 arachno1999 Accepted Answer ✓
Some thoughts against it: Chatbots are a crowded market already. Chatbots that are delivering value for a company are very product specific -> lot of training needed As the customers seem not to understand the process of building a chatbot, it will be very difficult to find the 10-20% of whatever. Selling cheap is not less tedious than selling high end.

To proof me wrong, just try to sell the idea as a not yet existent or as a fake product. Build a landing page where people can subscribe to get further information, lead a few prospects to this page and count the subscriptions.

👤 webmaven
I'm not familiar with the market, but it sounds viable.

You seem to have missed one trick though: offering a 10-20% discount in return for adding their (labelled) data to the common training corpus.

👤 brudgers
Its viability is directly proportional to the emphasis on sales instead of technology.

Business is not based on making good stuff.

Business is based on selling stuff.

Or to put it another way, technical development is the inverse of a con.

Good luck.