How can virtual agents have human-like conversations?

Aiden.ai’s approach to the NLP challenge

There are several different kinds of conversations that two humans can have. A passionate argument, an inquiry, or exchanging directions. We don’t think about the different complexities involved with these various types because it is completely natural for us. We also take for granted the intuition and context that make these conversations possible.

Recreating conversation, in all of its complex formats, is one of the most exciting modern challenges developers are yet to solve.

Understanding the meaning, “Natural Language Processing”, is the first step.

But we also need to understand “context”, and even model “intuition” to be able to develop a true virtual assistant.

How can it be done?

This is the question that everyone is racing to answer. Google, Facebook, Microsoft, scientists and developers around the world are working hard to figure out how a computer program can converse like a real human would. At Aiden.ai, we love a technical challenge — so naturally, we have joined the race.

Let’s look at a couple different types of conversations that we are developing into Aiden, our virtual colleague for performance driven marketers.

We’ll start with the easy one — reactive conversations

We consider a reactive conversation with Aiden to consist of the user asking a question and expecting an answer.

For example, the user might ask:

“How many paid installs did I get last week?”

From that question, Aiden can first determine the “intent” of the question. In this case, it is a metric query — the intent is to understand a specific number. Aiden then must decipher the important tokens in the sentence in order to figure out what number the user is asking about. Aiden identifies “paid installs” and a length of time “last week.” Based on these two entities and the intent of the question, Aiden is able to search the data he has access to and deliver an answer.

This is an example of basic NLP and is basically solved from a scientific development standpoint.

It becomes tricky when you want to add context. A basic example is the follow-on question, after the first question is asked by the user:

User: How many installs did I get last week?

Aiden: 1000 installs (from Sept 25th to Oct 1st)

User: What about last month?

At this stage, Aiden needs to remember what has been asked previously and to understand that the two questions are linked, in order to determine that this is another metric query, based on a different time frame.

Follow-up questions are a key element to build an efficient virtual colleague, as you would never repeat exactly the same sentences to ask for details on a specific topic.

With proactive conversations, it starts to get harder.

Here, the intention is for Aiden to identify interesting data points and proactively alert the user while making suggestions about how to interpret and take action.

For example, Aiden should be able to identify when a campaign has become “saturated” and suggest to its user that they pause or lower the budget. These types of proactive suggestions are exactly what you would expect from a living breathing analyst, so it makes sense to expect this behavior from a virtual colleague. In order to pull this off, it is critical that Aiden is equipped with machine learning capabilities so that he can learn what the user is interested in and find more data points that will be useful.

Our conviction, is that Aiden will become your virtual colleague.

Aiden is a conversational tool fully capable of speaking and answering questions in plain English. He might need a little help in the beginning but eventually he should be able to decipher context and background information well enough to answer any question posed to him correctly.

Additionally, he is be able to monitor marketing campaigns and assist the user with identifying anomalies that could impact budget, performance, and ROI. The goal is that Aiden will augment your data analysis by performing the mundane and time consuming tasks that often hold marketers back from answering more complex questions.

Sounds easy right? 😉 The truth is this is an extremely cutting-edge and exciting challenge that keeps changing as new developments are achieved around the world.

Want to join in on the fun? Aiden.ai is hiring. Check out our job postings and reach out to us if you are excited by a challenge!