When talking to Daneel in order to find a result following specific criteria in an E-commerce context, it is able to have a discussion with us, guide us towards better result through disambiguation and help us when we make mistakes, but how does it manages to do that ?
Experience leads to knowledge
In a previous article Bringing the power of cognitive services to my personal documents, apps and devices, we introduced you to the way Daneel stores its data.
In the memory graph, each node is a symbol, and those nodes are linked together when the meaning of their respective symbols tends to be semantically connected. For example the symbols represented by the words “red” and “color” will be often encountered together.
Now, the interesting part in the way symbols interact with each others is that the more Daneel encounters the same connected symbols, the stronger the link between the nodes will become allowing us to create more knowledge based on an abstraction of nodes combination.
It means that like you and I, Daneel is able to learn by practice, through the experience it has with symbols. It can then use its new knowledge to learn more and understand you better.
Use semantics for a better data comprehension
Armed with its new knowledge, Daneel can better understand the links that symbols can have between each other.
A use case of this skill appears when, in an E-commerce context, Daneel faces the memorization of a product with poor specifications. Based on previous experience, it will be able to create information from other sources like text descriptions, images or even similar products.
To quote our previous example, if Daneel recognize the symbol “red” in the description of the product, it will know that “red” is strongly connected to the “color” symbol and in this precise context, it will enhance the product specification on its own, Daneel use its knowledge for a better data comprehension.
The same mechanism is used, to help the user making a choice when the search sentence contains only one word: “apple”. In this case, Daneel will start a conversation to guide the user in his search, skipping unwanted results and keeping track of the context of the search using semantics. Indeed, in the video, we can see Daneel asking Anthony if he wants result for the “category” , the “title” or the “color” referring to the “apple” word.
In a near future
These use cases of Daneel knowledge application are pretty basic and the possibilities are quasi endless, moreover, the API opened to developers will allow them to add new behaviors to the cognitive system we are working on as you can read in the Application development to development of behavior article.
It is completely feasible that in the near future Daneel will automatically enhance data from images like extracting the color of the main subject on the image, or classify a song from the analysis of its tempo and the instruments playing.