Boosting document analysis with NLP
CNH Industrial chooses a neural network-based advanced analytics solution to improve extracting key insights from the dealer comments obtained from warranty claims.
#NLP
#Neural Network
#Data Analysis
)
Scenario
Manually analysing claim comments
CNH Industrial is a global manufacturer of agriculture and construction equipment. The company has more than 37,000 employees, and operates in 180 countries. In order to determine the main causes of the failure of their equipment, CNH Industrial uses a set of documents - referred to as claim comments - filled in by mechanics during repair, which describes a customer’s complaint (including the cause of the machine’s failure and the implemented solution). Engineers had to read each single claim comment to identify the most common issues.
The challenge
Deepening the document investigation
The claims’ analysis was conducted one-dimensionally, by looking at the product category and filtering by single word, thus limiting the depth of the investigation. Analysis was also constrained by the engineers’ human limitations - e.g., ability of keeping track of failure mode or the number of documents they could process at once.
Hence, the challenge of this project was to make the analytical process more scalable and effective by introducing higher in-depth analysis.
How we did it
A neural network solution
The problem was tackled in an innovative way: Target Reply opted for a semantic approach instead of the classic frequentist one, not limiting the analysis on single words probabilities but relying on the whole semantic context.
Holistic Analysis
An NLP algorithm introduced a new dimension aimed at identifying the topic of each claim, fostering a cross-product analysis previously unattainable.
Maintenance ability
The solution was built in modules, so that each component was easily upgradable whenever new and innovative algorithms were released.
Dashboard integration
Results were encapsulated in a dashboard that users could easily explore and customise.
The results
Fast and meaningful insights
The developed algorithm manages a large amount of raw text and provides fast and meaningful insights. The outcome is delivered through a dashboard that is able to easily convey the model’s complexity in an interactive platform. Through the dashboard, users are able to analyse topics shared by different claims promoting a more efficient and effective analysis. The reusability of the algorithms grants flexibility to the process, which is now able to adapt to different contexts and databases.
CNH Industrial (NYSE: CNHI / MI: CNHI) is a world-class equipment and services company. Driven by its purpose of Breaking New Ground, which centers on Innovation, Sustainability and Productivity, the Company provides the strategic direction, R&D capabilities, and investments that enable the success of its global and regional Brands. Globally, Case IH and New Holland Agriculture supply 360° agriculture applications from machines to implements and the digital technologies that enhance them; and CASE and New Holland Construction Equipment deliver a full lineup of construction products that make the industry more productive. As a truly global company, CNH Industrial’s 37,000+ employees form part of a diverse and inclusive workplace, focused on empowering customers to grow, and build, a better world. For more information and the latest financial and sustainability reports visit: cnhindustrial.com
Target Reply is the Reply Group company that specialises in implementing solutions involving big data and advanced analytics. Target Reply supports organisations in identifying their needs for solution design and implementation by using technologies for data integration, data modelling and predictive analysis. In the process, it utilises the most innovative tools available in the business discovery and big data fields. Target Reply has gained significant experience by working with Italy’s major business’ challenges and is able to operate in all key markets, including telecoms, finance and manufacturing.