Industry
Retail
- Manual effort for Subject Matter Experts (SMEs) and external contractors was significantly reduced, leading to saving of approximately $1.3 million per year.
- Legal costs due to non-compliance which ranging between $1-$20 million depending on the severity of the case was also saved
- The solution was more than 87% accurate and was able to constantly improve, learn and generalize new types of input documents. This will have positive year-on-year impact.
The R&D team of an American global food manufacturer was tasked with manual scanning of scientific literature and news commentaries to ensure timely awareness of food safety and nutrition issues while assuring regulatory compliance for effective change management. Quarterly reports were being analysed by R&D associates, SMEs and certain commercial entities and external contractors.
The objective was to increase volume of documents scanned and reduce the time spent on categorization while automatically rejecting irrelevant abstracts without manual review.
- A vector space i.e., map of discrete words was created using Recurrent Neural Networks wherein words similar in context appear closer in the map. The embeddings also infer the meaning of words as per the context of their appearance.
- A Natural Language Processing (NLP) based tool was designed for the SMEs to download input docs and accept/reject/review results to enable automatic rejection of irrelevant topics.
- Relevant of documents was determined via relationships between authors, journals, citations. The same as extracted through NLP and visualized in Document Vectors, Author Relationship Networks and Reference Networks.
