Most of us are regularly inundated with ads and it can be increasingly difficult to get folks attention. At the same time, people are increasingly accustomed to immediate responses on mediums like SMS, Facebook chat, and Twitter. As a result, conversational interfaces like chatbots are becoming increasingly popular for applications like customer support, product recommendations, targeted promotions and product feedback aggregation.
What almost all conversational interfaces have in common are problems with machine reading comprehension. Leveraging our team's more than 60 years of combined deep NLP experience, we've helped a number of companies improve their conversational understanding capabilities so their bots are capable of deeper and more effective conversations. In one case, we were able to efficiently extend one conversational domain from the recognition of just a handful of response categories, to over 50, all with very high accuracy. In another case, we were able to leverage modern LSTM deep learning neural nets (i.e. AI) to outperform more traditional SVM style learning techniques. In another case, we were able to, in just a few short days, improve net effectiveness of algorithms by a significant percentage through the use of deeper semantic grammar based model features to the thrill of the client's team. In all cases, off the shelf alternatives were entirely insufficient as these models are typically trained on very different data and have no means for heavily customizing conversations.