At Crimecheck, we continuously look at ways to leverage tech to increase value to our clients – through improvement accuracy, faster results or unlocking new possibilities. Machine Learning has been one of our strong areas, using our huge data moat to keep improving our products. We keep adding trained models to step in for humans to reduce their effort and errors – like filtering out irrelevant cases, automating lot of manual tagging of data, marking risk categories based on legal categorization of the case, etc.
Recently we have launched a Deep Learning model to recognize whether a given name is a company or an individual. Trivial as it sounds, it is very hard to get a high accuracy on this, given that lot of companies have person’s names, and given typos / missing qualifiers, names in regional languages etc. And it has a huge impact on search results; just knowing whether a given court record involves a person or a company helps us narrow down our search pool significantly. Our team went into research mode to come up with this Deep Learning model based on LSTM, which they methodically compared with other approaches like Classical machine-learning model, and determined to be better. If you are tech-oriented and want to understand how we achieved this, head over to our Medium page with full details.
We are happy to share the value with our clients, and the knowledge with the whole community 🙂