Arenji began as an evaluation of the state of the art of Natural Language Processing (NLP) services, with the intention of applying these to knowledge retrieval for manufacturing applications. Curious about the hype but skeptical of the promises of the IBMs Watson ads, we wanted to see what was really possible. There’s been a lot of excitement about chatbots and AI in the past few years and we wanted to separate the reality from the hype… And there is a lot of hype out there – bots are nowhere near how they’re portrayed in the ads. That said, there have been great strides in the machine learning/NLP space that provide a foundation to building natural language based user interfaces. The key is to control the domain scope of interaction.
Intent prediction and named entity extraction are two of the common machine learning / NLP models that have been employed successfully. Command and control type applications such as Alexa, Siri, Cortana etc. are able to identify one of the many ways someone might ask to play a song, turn on the lights, etc. And they work well for limited domains and general requests. Asking about the weather for today/tomorrow works well, whereas asking if we’ll see the sun sometime this week doesn’t usually yield the expected result. Some of the limitations are rooted in the sheer number of permutations of what someone might ask these systems and that they need to be trained to handle them. Other limitations are rooted in how complex a pattern is to recognize in the phrasing.
Building chatbots is fun and rewarding. It’s a new space, and it’s easy to project that you’re interacting with a human or sentient computer. While many low level services and bot frameworks exist, we found that building a bot still had a steep learning curve and required a deep background in computer science. We set out to make building a bot something similar to creating a survey. It’s probably no surprise then that our first product offering is focused on Market Research.
There’s no substitute to having a conversation. It’s the best way to get to the ‘why’, to understand the customer’s frame of reference. Our chat based surveys allow for the surveyor to hear the respondent in their own words yet get the benefit of an automated system to quantify the responses. If you’re interested in trying it out we have a beta version online. (Just click here to sign up)