How Entity Extraction Helps Manage Third-Party Risk in the Shipping Industry

Shipping Companies Need to Know Their Partners Well

A serious challenge for the shipping industry is figuring out who it is doing business with. As supply chains become longer and more complex, and as bad actors obscure and misrepresent their identities, it is increasingly difficult to get a clear and detailed picture of business partners. A well-known example is that offered by Iranian companies. The Islamic Republic of Iran Shipping Lines (IRISL) uses many deceptive and fraudulent practices to conceal its identity and circumvent the international sanctions imposed on it: it changes names and ownership of vessels and even repaints them; it has also set up front companies outside Iran and then assigned vessel ownership to them.

The Requirement to Perform Due Diligence

In order to reduce the chances of getting mixed up with bad actors, shipping companies are subject to regulation. They must conduct what’s called “third party” due diligence of potential business partners in the areas of anti-bribery and corruption, anti-money laundering (AML), and trade sanctions. The last few years have witnessed an increasing emphasis by US and EU governments on enforcement of these issues. In 2019, two shipping companies were fined for violations of U.S. Federal law. In addition, much like Iran, a rogue nation such as North Korea is well known for elaborate schemes to avoid sanctions, including false flagging, ship-to-ship transfers, setting up of fake shipping companies, and other hallmarks of fraudulent behavior. Legitimate companies need to steer clear of all these types of compromising situations.

Using Adverse Media Monitoring to Vet Business Partners

An important tool to defend against such third-party fraud is Adverse Media Monitoring. Adverse media is any kind of negative information about a person or organization discovered across various sources such as news media, blogs, social media, and internet forums.

One advantage of monitoring media in the due diligence process is that it is completely up-to-date, as opposed to the entries in sanctions lists, which will be inevitably somewhat behind the times even if they are updated regularly. This matters when a potential partner was recently indicted and the court proceedings have not begun or are in progress – they won’t be on any watch list yet.

Why Adverse Media Monitoring is Challenging

Traditional search engines such as Google would appear to be a good tool for performing Adverse Media Monitoring, but they contain one major Achilles’ Heel: they rely exclusively on keyword search, which is not adequate when searching on items such as company names.

For example, company names are frequently ambiguous:

• Apple (company) vs. apple (fruit)
• Amazon (company) vs. Amazon (river)
• Alphabet (company) vs. alphabet (abstract common noun)

Just searching on “apple” will generate too many false positives. Trying to narrow the search by, say, searching for “Apple, Inc.” will miss the many valid examples of “apple” referring to the company, hence more false negatives.

Besides the inherent limitations of keyword search, human staff currently review on-line data sources looking for adverse material. Unfortunately, no purely human approach will scale to meet the requirement of at-scale monitoring of extremely large amounts of unstructured text data. A more automated approach is required.

How Entity Extraction Enables Adverse Media Monitoring

An important AI technology that supports Adverse Media Monitoring is Entity Extraction. The latter enables discovery of adverse information from a massive volume of unstructured text data as found in news media and other sources.  In the search for adverse information regarding third-party risk, companies need to collect information on an ongoing basis to derive maximum benefit.

Entity Extraction automatically identifies semantic concepts in text such as named entities, e.g., “Cartwheel Shipping Company.” Entity Extraction doesn’t just work from a list of names. Its algorithms are capable of dynamic name recognition: they use contextual clues and other linguistic information to identify instances of names that have not been seen by the software before. This is a powerful way of ensuring that coverage is complete.

Advanced Entity Extraction goes beyond just named entities and identifies more complex semantic concepts such as relationships and events. Event Extraction is particularly valuable in identifying adverse information about third parties because it identifies questionable or criminal events in which the third party has been involved.

For example, a sentence such as the following contains three types of adverse event information about a company:

• “A federal grand jury has indicted Chartworld Shipping Corporation with bribery and money laundering.”

The output of Event Extraction would be:

• • Event: Indict
  • Entity: A federal grand jury
  • Party Indicted: Chartworld Shipping Corporation

• • Event: Bribery
  • Briber: Chartworld Shipping Corporation

• • Event: Launder Money
  • Launderer: Chartworld Shipping Corporation

Another example:

• “Cimpship Transportes Maritimos S.A. was hit with a $5M fine Friday.

• • Event: Fine
  • Party Fined: Cimpship Transportes Maritimos S.A.
  • Money: $5M
  • Time: Friday

Some of the key strengths of Event Extraction are:

• It finds specific adverse event types regardless of the words used to convey them (e.g., indict, indicted, indictment)
• It assigns specific roles to each entity involved (e.g., entity that indicts vs. entity that is indicted)

Event Extraction makes it possible for natural language data to be searched more like a conventional database search, which minimizes the problems of false positives and negatives discussed above.

Summary

In sum, Entity Extraction provides sophisticated Adverse Media Monitoring to identify the third parties in shipping who are engaged in corrupt activities. It gives an organization the capability of monitoring a wide range of media more accurately and efficiently with much less use of manual review.

In this age of lengthy and complex supply chains, Entity Extraction makes them a much less risky environment for companies who depend on them.