This document proposes a framework for question answering over pattern-based user models. It combines a frame-based representation of natural language questions with a context-aware, graph-based approach to knowledge extraction. The framework performs semantic analysis of questions, identifies relevant concepts in ontologies, constructs local contexts around those concepts, and ranks contexts to generate answers by matching to the question analysis. An example demonstrates applying the framework to answer the question "How often does Ann like to drink coffee?". The framework aims to support question answering over complex conceptual models and user profiles.