Predicting Interest: Another Use for Latent Semantic Analysis

Abstract

Latent Semantic Analysis (LSA) is a statistical technique for extracting semantic information from text corpora. LSA has been used with success to automatically grade student essays (Intelligent Essay Scoring), model human language learning, and model language comprehension. We examine how LSA may help to predict a reader’s interest in a selection of news articles, based on their reported interest for other articles. The initial results are encouraging. LSA (using default corpus and setup) can closely match human preferences, with RMSE values as low as 2.09 (human ratings being on a scale of 1-10). Additionally, an Adapting Measure (best parameters for each individual) produced significantly better results, RMSE = 1.79. Keywords: Adapting Measure; Latent Semantic Analysis; LSA; human interest prediction; predicting ratings; news articles


Back to Table of Contents