1. AX Chatbot - what is it and what does it do?
Our visionaries team came up with a chatbot application, that formulates it’s answers using an AX Semantics NLG application. At the moment, you can ask this bot for ceratin activities (Can I do a certain sport today? Can I take the bike to work? Can I throw something on the grill?) . The bot will reason what will be the ideal weather conditions to do that activity an tell you how the current weather situation fits your plans.
(“Yes, you can do that because today is going to be a sunny day. Keep in mind that there will be periods with clouded sky though.”)
There are negotiations with four early adopter customers ongoing at the moment.
2. Benefits in general
In contrast to a regular chatbot, which understands questions to a certain extent and selects relatively unflexible answers, an NLG powered chatbot can answer in a lot more detail and with a lot more respect to the user’s context. That shows when the bot picks up elements from the question or when it draws it’s own conclusions from the topical data and the user’s data.
3. Common Use Cases
- Serviceline chatbot: A bot is given access to customer account data and trained to answer and prequalify typical support questions until a service line operator has time to pick up the conversation.
- Product guide chatbot: A bot is given access to a retailer’s PIM system and optionally customer account data and can consult and guide the customer through his purchase.
- Service content bot for publishing: A bot is given access to a data source of general interest (air quality, weather, sports, traffic) and offered to readers as an additional service.
- BI / IoT bot: A chatbot is given access to a live set of sensor- or business process data and provided with an analytics backend (i.e. azure machine learning). The bot is then trained to alert users when something critical happens but also to summarize, assess and explain what has happened in the data when the user asks.
4. How it was done
The chatbot consists of four main components:
- The NLG component, in our case AX Semantics, that generates the actual answers.
- The Dialogue planner, in our case API.AI, which understands the user’ questions and provides intentions and entities from them.
- The action frontend, in our case Google Actions for Google Assistant and Google Home, which controls the channel over which the user interacts with the bot. For our prototype, Skype, Slack, Telegram, Alexa and Facebook Messeger can be realized, but with additional effort.
- A data source which gives the bot knowledge about the topic. Since it must be addressed at runtime, it will probably always include something similar to nr5.