Morph.ai has it's own AI (Artificial Intelligence) Engine & also supports integration with API.ai.
An AI engine can let your bot do the following:
- Parse & understand user sentence: The AI engine will use NLP techniques to understand the user sentence.
- Extract meaningful information: It also extracts all the meaningful information from the sentence using NER techniques.
- Map it to a supported intent: Then the AI engine uses classification algorithms to find the user intent.
- Use this information to reply back: Then the bot can use the identified intent and extract information to give a intelligence response to the user.
You should use Morph.ai's AI engine because:
- It's as good as API.ai
- It's 10 times cheaper then API.ai
- You get more control and can tune the AI model better
- Supports real time training
- Support mostly all the languages out-of-the-box
- Very easy to debug. We also keep a complete trace of any understanding done by the system which can help you find why any result came and to which training data it was mapped to.
- Has support for
*(Asterisk) entity which can match anything.
- We can easily do business specific customizations.
If you are already using API.ai then we suggest to keep using that in the start. First migrate your bot on Morph.ai platform and then gradually migrate your intents and entities from Api.ai
It's as par with API.ai. We have benchmarked our AI Engine with API.ai. To get a detailed report please contact us.
We have written our own AI engine after trying out all the available AI engines and finding something always missing in them. Our AI engine is specific to only chatbots and hence provide you with all you need from an AI engine to make chatbots. Also we wanted to reduce the cost of using AI and NLP techniques in chatbots.
Following are some of the techniques we use:
Morph.ai's AI Engine works in multiple phases. Instead of creating a single AI model, we have created multiple AI model each one responsible for single intelligence. This helps us reusing the same model again and reducing the cost of an training event.
We have developed an scoring algorithm to determine the confidence of the understanding. This algorithms help the client to understand how confident is the AI engine in understanding any sentence.
To make our scoring more suitable to chatbots, we have added concept of penalty in the scoring algorithm. It take care of the un understood sentence. This ensure you get minimum false positives.
Because of having distributed model, we have can easily replace a single model with any other model. This helped us grouping language specific features in isolated models. We have also developed a generic model which is dependent bag of keywords model independent of the language. This lets you use our AI engine for mostly all the languages. Find more on this here. This model
Our AI engine is completely configurable and you can train any part of it. This gives us the power to support multiple training methods.