The analytics behind chatbots

Facebook says tens of thousands of developers are working on chatbots. Microsoft, Google, Amazon- almost all major companies have indicated their interest in chatbots.

Machine Learning and Analytics to the fore-

Analytics India Magazine states that some of the deep learning techniques that are used to build chatbots are based on either Retrieval models, which use predefined responses or Generative models which generate new responses from scratch.

While retrieval models don’t make grammatical mistakes, as they have a database of responses to choose from; they are unable to handle cases for which no appropriate predefined response exists. On the other hand, Generative models are hard to train and are more likely to make grammatical mistakes, even though they are ‘smarter’ as they use machine translation techniques to translate from an input to output.

Facebook Messenger bot engine will be following the retrieval model for most companies as the starting point of a bot would be a set of interaction rules defined by the developer as some basic guidelines on handling inquiries. Further, the application will start amassing chat logs, and soon it will be able to take advantage of the tool’s machine learning component to scan the information for recurring communications patterns and factor them into its programming.

Initial experiences show that while chatbots might be the future of personalized communication between brands and consumers, and have massive potential as the de-facto data collection tool,  there is still a long way to go. At the same time, they could also be a very dangerous threat to a customer’s online privacy & security over time. As they are supposed to behave as human as possible by mimicking the language, they may end up collecting sensitive or confidential information. The problem is that this data will get stored somewhere which could be misused.

Have you thought about developing, or are in the process of developing chatbots? Do share your experiences!

Share This Post

Leave a Reply