In articles, the identifiers of person/organization like name, address, DOB, occupation, registration/tax/govt IDs, etc. are matched along with the scores, based on the number and combination of identifiers. Techniques like named entity recognition, parts of speech tagging, pattern-based matching, etc.are applied.
This process establishes the relationship between the identifiers and determines if the article is really talking about the entity or is it just mentioned for some other reason. It extracts sentences from the article which talk specifically about the relationships or illegal activities using trained ML models using SVM, neural networks and NLP techniques like co-reference resolution, dependency parsing, etc.
Negative/contextual words like arrest, money laundering, cheat, etc. are matched and articles are marked into high/medium/low risk, based on the matches.
Determining the sentiment of an article, whether positive or negative, using pre-trained or custom-trained sentiment models help in discounting as it gives an insight of how negative an article is.
Application of these four checks on adverse media articles and taking a weighted average score while discounting can give a complete picture on whether the article is important for screening or not. Required evidences and case reports can be generated during the process.
Advantages of name screen discounting
Irrelevant articles can be discounted, avoiding the need to analyse false-positives generated while collecting adverse news.
- Automation using NLP and ML saves effort and reduces time by nearly 60% as 1-2 hours of manual processing time is brought down to 20-40 minutes (See Figure 3).
- Annotated articles with highlighted keywords, identifiers and important sentences can be generated as part of the process which can serve as evidences in later stages
- .It can be helpful in creating case reports and audit trails as part of the screening process in an automated way, thereby eliminating human bias.
- Feasibility of prioritizing any of the four perspectives for discounting by giving high weight-age according to respective organization's regulatory policies.