A scalable maritime platform providing services through an NLP-based intelligent personal assistant

A scalable maritime platform providing services through an NLP-based intelligent personal assistant

In an era characterized by rapid technological evolution and massive data acquisition, humans find it progressively harder to compete with the increasing complexity of handling immense volumes of information. In maritime industry especially, employees face huge amounts of data that need to be processed in order to extract meaningful inference for vessels’ efficiency every day, striving to compete with these challenges. Advances in Artificial Intelligence and Computer Science are however able to lift this weight up, applying high computational power and smart, effective algorithms to organize, process and visualize tons of useful, yet hard to cope with, data.

To this scope, we focused on maritime and implemented a conversational personal assistant specifically designed for this field. Our chat bot is able to recognize messages directly in natural language, parse them and extract all the necessary information, to finally call special agents to create an answer and send it to the users’ email, Skype or any other channel of their choice. Moreover, users have the ability to provide information to our system for future use, like setting the next port of their vessel.

The strong points of our system are its scalability and its ability to analyze the users’ messages in multiple levels, using Natural Language Processing techniques. For this purpose, a machine learning classifier is trained with several messages to learn what a message is about, from a predefined set of classes that cover the core of interest of maritime employers and employees. New categories can be added with minimum effort, as our classifier models are fully retrainable with more classes and messages. When a user sends a message to our system, linguistic analysis is applied in two phases. First, the classifier finds the topics of the message, i.e. the classes it belongs. During this phase, the message is subjected to multiple preprocessing stages and transformations. Next, given the message categories, special NLP parsers analyze the message to extract the exact parameters given or requested, like dates, quantities, filters etc., in order to respond precisely to the original message.

Two-stage analysis with a generic classifier and special parsers allows for effective handling of a wide variety of messages, from as simple as “What is the speed of vessel?” to as complex as “Get avg fuel consumption of diesel generator 2 from 6/7/2017 to 20 days after last Christmas, only when main engine speed was above 35RPM”. Extended experiments (and our customers’ satisfaction!) prove the quality and the importance of our system’s results.

Book your ticket now

Do you like this session? Join 600+ attendees by registering now and live the Voxxed Athens experience