Forecasting of Trends in Legal Spend Management

摘要

The paper describes a framework for forecasting narrative trends (text-based description of cost items) in legal spending. This is based on the application of topic discovery and time series forecasting. The algorithm presented in this paper discovers a number of abstract topics in a corpus based on clusters of words that are found in each line item spending document, along with the respective frequency of those words. Specifically, Latent Semantic Analysis transforms a sequence of cost descriptions into a set of numerical Topic-based univariate time series. The resulting set of time series is used to forecast future trends using the ARIMA (AutoRegressive Integrated Moving Average) approach. This type of semantic forecasting of spending trends can facilitate the discovery of counterparty intent(s) and proactively adjust the litigation strategy (prove/disapprove a claim, counterclaim, etc.).