Teaching a computer to read human language and comprehend meaning has been one of Artificial Intelligence’s holiest grails.
Reading comprehension is challenging – even for human readers. In order to do it well, a reader not only needs to process a significant amount of reading material, but also needs to develop the capability to interpret questions correctly and find answers accurately. Surprisingly, computers are performing the task relatively well.
A Brief Review of Natural Language Processing
Early research into reading comprehension started in the late 1950s and gradually evolved into the subject of Natural Language Processing (NLP) – also known as Computational Linguistics. NLP leverages a combination of Machine Learning (ML), Artificial Intelligence (AI), and Linguistics techniques to comprehend, interpret, and even generate human language information.
Stanford University spearheaded the research by creating the Stanford University’s Question Answering Dataset in 2016. This dataset is a collection of paragraphs taken from more than 500 Wikipedia pages spanning a wide range of subjects. The dataset also provides for 100,000+ questions. The answer to each question is a segment of text from a reading passage. Researchers built intelligent algorithms to “understand” both the questions and the passages, then propose the best section in the passages to answer each question.
A typical passage and questions look like this:
In meteorology, precipitation is any product of the condensation of atmospheric water vapor that falls under gravity. The main forms of precipitation include drizzle, rain, sleet, snow, graupel and hail… Precipitation forms as smaller droplets coalesce via collision with other raindrops or ice crystals within a cloud. Short, intense periods of rain in scattered locations are called “showers”.
Question 1: What causes precipitation to fall?
Question 2: What is another main form of precipitation besides drizzle, rain, snow, sleet and hail?
Question 3: Where do water droplets collide with ice crystals to form precipitation?
within a cloud
Researchers around the world have been attracted to develop and test question answering algorithms against this dataset. As a result, the leaderboard of the Stanford Question Answering Challenge has grown into one of the world’s top battlegrounds for AI research.
As of June 2018, the #1 algorithm on the leaderboard has reached 83.88% accuracy, already exceeding average human performance (82.30%).
This achievement has led to Round 2 of the battle. Currently, no matter whether an input paragraph is relevant or not, algorithms will attempt to produce some answer for the questions given. This has propelled researchers to tackle the next challenge – being able to identify when information is insufficient and say, “Sorry, I don’t know.”
Although there’s still some debate over the fairness of these challenges – whether they are set up in a way that favors the machine, and whether the machine has truly outperformed humans or not – it’s intriguing to see how fast the forefront of AI research reaches new heights.
Applications and Possibilities in Supply Chain Management
In regards to supply chain management, text-based Machine Learning techniques can inspire many use cases in Supply Chain Management.
- Intelligent Product Classification
By running analysis over product names and descriptions, retail companies will be able to streamline the process of acquiring new merchandise, categorizing them, and then populating them on the correct pages of their online catalog
- Trade Compliance
Accurate product classification is a compliance requirement when conducting international trade. Being able to decipher the human language as it relates to product information and tariff schedules – as well as being able to identify the accurate product category and recommend tariff codes will save enormous amount of time for companies.
- Manual Descriptions Integrated with Chatbots
Text descriptions entered manually by users would be a valuable supplement to system data and would deliver a more holistic picture of a supply chain process. Take shipping as an example: If various parties recorded information such as reasons of delay or other anomalies – if Natural Language Processing capability were embedded into a system – users would be able to avoid reading through “Remarks” or “Notes.” When connected with chatbots, a user would only need ask, “Why is this shipment late?” A chatbot could gather the relevant information, analyze it, and respond in human language
- Business Intelligence
With analysis running regularly on unstructured text data, businesses would be able to utilize aggregated results as new data streams and feed them into Business Intelligence tools for monitoring and reporting. More business patterns could be uncovered that benefit organizations.
AI and machine learning tools and other emerging technologies are growing at a rapid clip in the Oracle SCM Ecosystem. Check out a few of our tools and platforms available now.