Introduction to Conversational AI

Original article was published on Deep Learning on Medium

One day when I got my daily morning coffee at the train station, I overheard this conversation between the service staff and a customer:

  • Service: Good morning, what would you like?
  • Customer: A coffee, please.
  • Service: Would you like a donut with that?
  • Customer: A donut is a great idea.
  • Service: What was that again, please?
  • Customer: I would like to have a donut with a cup of coffee.

Well, that conversation used up more than the normal running time. What happened? Why did the service staff ask again what the customer wanted? The service person — just like you and me — would have reacted best to the most simple of answers, yes or no.

And this is exactly how Conversational AI computers function. The difficulty lies in recognizing patterns from a text, identifying its context and fetching all results because sometimes the information is not clear.

What is Conversational AI?

Conversational AI is a set of technologies behind automated messaging and speech-enabled applications that offer human-like interactions between computers and humans. This is any machine that a person can “talk” to. It can also discover and filter specific words in texts. (Source: https://www.interactions.com/conversational-ai/).

The picture below is a demonstration related to N26 Bank to support their customers ahead of time. It is so to say the initial contact phase before the customer is redirected to the real, man-powered customer service. It significantly saves the cost of customer service.

source: https://rasa.com/

The Principles of Conversational AI

If you would like to implement Conversational AI in your business, you must consider the following aspects:

  • Objectives and context: Begin by defining the purpose of conversational interactions in your business.
  • Security and privacy: These are major concerns when it comes to bots. Almost half of the users are concerned about safety.

Conversational Use Cases

  1. Connection Business with Customers: E-Commerce and Customer Services

2. Connection Business with Employees: Feedback, Helpdesk, On-boarding

3. IOT: Home entertainment

4. Text Analysis:

  • Text QA: Question/Answering Component
  • ODQA: Open Domain Question Answering: answers any question based on the document collection, covering a wide range of topics.
  • Ranking: The ranking component solves the tasks of ranking and paraphrases identification based on siamese neural networks with integrated semantic similarity measures.
  • Entity recognition: Named Entity Recognition (NER) classifies tokens in text into predefined categories (tags), such as person names, quantity expressions, percentage expressions, names of locations and organizations, as well as expression of time, currency and others.
  • Intent classification: Intent classification recognizes intents based on user utterance.
  • Insult detection: That predicts whether a comment posted during a public discussion is considered insulting to one of the participants.

Tools and Framework

DeepPavlov

– DeepPavlov is an open-source conversational AI library built on TensorFlow and Keras (http://deeppavlov.ai).

SAPConversationalAI

– SAP-Conversational AI is to be found here: https://www.sap.com/germany/products/conversational-ai.html.

Rasa

Perspectiveapi

– You will get more information on Perspectiveapi here: perspectiveapi.com.