Abstract:Modeling human behavioral data is challenging due to its scale, sparseness (few observations per individual), heterogeneity (differently behaving individuals), and class imbalance (few observations of the outcome of interest). An additional challenge is learning an interpretable model that not only accurately predicts outcomes, but also identifies important factors associated with a given behavior. To address these challenges, we describe a statistical approach to modeling behavioral data called the structured sum-of-squares decomposition (S3D). The algorithm, which is inspired by decision trees, selects important features that collectively explain the variation of the outcome, quantifies correlations between the features, and partitions the subspace of important features into smaller, more homogeneous blocks that correspond to similarly-behaving subgroups within the population. This partitioned subspace allows us to predict and analyze the behavior of the outcome variable both statistically and visually, giving a medium to examine the effect of various features and to create explainable predictions. We apply S3D to learn models of online activity from large-scale data collected from diverse sites, such as Stack Exchange, Khan Academy, Twitter, Duolingo, and Digg. We show that S3D creates parsimonious models that can predict outcomes in the held-out data at levels comparable to state-of-the-art approaches, but in addition, produces interpretable models that provide insights into behaviors. This is important for informing strategies aimed at changing behavior, designing social systems, but also for explaining predictions, a critical step towards minimizing algorithmic bias.

Unsupervised Search-based Structured Prediction

Search Action Sequence Modeling With Long Short-Term Memory For Search Task Success Evaluation

Search Task Success Evaluation by Exploiting Multi-View Active Semi-Supervised Learning.

Discriminative unsupervised learning of structured predictors

Search and learning for unsupervised text generation

Unsupervised Learning of Discourse Structures using a Tree Autoencoder

Learning Output Embeddings in Structured Prediction

Predicting Above-Sentence Discourse Structure Using Distant Supervision from Topic Segmentation

Efficient Decomposed Learning for Structured Prediction

Going "Deeper": Structured Sememe Prediction via Transformer with Tree Attention

SpeedMachines: Anytime Structured Prediction

The structure of verbal sequences analyzed with unsupervised learning techniques

A Neural Probabilistic Structured-Prediction Method for Transition-Based Natural Language Processing.

Methods and Theories for Large-scale Structured Prediction

Sequential Search with Off-Policy Reinforcement Learning

A Supervised Learning Approach to Entity Search

Data-efficient Active Learning for Structured Prediction with Partial Annotation and Self-Training

Semantics and Geography Aware Hierarchical Learning for Sequential Crime Prediction

Predicting and Explaining Behavioral Data with Structured Feature Space Decomposition

Learning Approximate Inference Networks for Structured Prediction

Unsupervised Inference of Data-Driven Discourse Structures using a Tree Auto-Encoder