To understand a little more about how our indispensable artificial intelligence for e-commerce works, you need to understand some topics such as: DeepFAQ, Knowledge Graph and ECQUAD.
And in this content we are going to demystify these terms so that you understand how we use each one of them in our artificial intelligence for e-commerce.
So, paper and pen in hand to write down everything that is most important and enjoy this super article developed by our artificial intelligence team.
DeepFAQ is a natural language processing technique that uses deep neural networks to answer questions based on a set of training data.
It is designed to learn how to extract information from a large structured and unstructured dataset and answer questions with high accuracy.
DeepFAQ is used in applications such as web search, virtual assistants and autoresponder systems.
GoBots, in partnership with the State University of Campinas (Unicamp), innovated by creating a tool that uses the customer's existing question and answer base to respond to new questions asked later.
This tool is a complete question and answer system based on similarity between questions.
Then the questions answered by the attendant are stored, with that a certain frequency of repetitive questions and answers begins to be created.
With this, DeepFAQ is able to determine that a new question fits this pattern and finally uses previously given answers to answer it.
Here below is an example of DeepFAQ working in practice:
It is interesting to point out that DeepFAQ uses exactly the answer previously given by the attendant, so it is necessary for the Artificial Intelligence team to add cleaning rules to these answers.
So it may happen that an answer given by DeepFAQ contains some word that these conditions could not clear, and a not so clear answer may be sent.
Knowledge Graph (KG)
Knowledge Graph is a knowledge representation system that uses a set of entities and relationships between them to represent information in a structured and interconnected way.
These entities can be real or abstract things like people, places, events or concepts, and relationships can describe how these entities are connected to each other.
The main idea is to represent knowledge in a way that can be easily understood and processed by machines.
Knowledge Graphs are widely used to support tasks such as information search and discovery, recommendation and question resolution.
For example, the Google uses a Knowledge Graph to present relevant information about people, places and things when someone does a web search.
O Facebook also uses Knowledge Graph to offer friends recommendations and relevant events.
Therefore, this is another tool, which also uses the responses previously given by the attendant.
But, it is worth mentioning that the Knowledge Graph differs a little from the operation of DeepFAQ.
Because in the case of usability here within GoBots, it is specific to the automotive industry, focusing on model compatibility questions and answers, such as:
- “Fits on corolla 2001”
- “It works on the 2010 civic”
- “It works for corsa 2001” …
Another interesting difference is how the Knowledge Graph works.
Because instead of analyzing the frequency of questions and answers, it ends up creating “triples” of knowledge, so when a question comes about the “corolla 2001” and the answer is something like “Yes it will/It won’t” (it doesn’t matter how it is writing) the Knowledge Graph is capable of creating a “box” of information. In the “information box” there would be data such as:
- The product where the question was asked;
- Which vehicle model was sent in the question;
- Whether compatibility is positive or negative.
That way, the next time a question is asked about this product, with the existing model in the box, the Knowledge Graph will pick up the registered compatibility answer.
It is a very advanced tool, which has a tendency of high hits.
Here below is an example of it in operation.
In the example above, the “little box” we made reference to is “Full Compatibility”.
The newest tool from GoBots in partnership with Unicamp, still in the testing phase.
This tool is very different from the previous ones because, instead of analyzing the history of questions, it will automatically analyze the description and characteristics of the product to find the most appropriate answer to the user's question.
So imagine a scenario where the user asks:
“How thick is the product?”
And this information is not found in the characteristics of the product, but it exists in the description.
In this case, there would be no possibility for the GoBots artificial intelligence to automatically respond (unless it was previously trained by the AI team).
That is why the ECQUAD tool was created, because even without prior training by the AI team, the artificial intelligence using this tool could find and send the answer to the user.
Description contains: Thickness: 4mm
Pergunta: “How thick is the product?”
Resposta: “Hello, the thickness is 4mm. Sincerely GoBots.”
Another example we can cite is the one below:
Now that you've seen all the technologies that are involved in our artificial intelligence, what are you waiting for to have it working for your e-commerce 24 hours a day?
Request a demo via WhatsApp by clicking here.
Did you like this article?
So don't forget to share with someone who will also enjoy reading this content.