Felix: Scaling Inference for Markov Logic with an Operator-based Approach
Feng Xing Niu, Ce Zhang, R Christopher, JUDE W. SHAVLIK · 2011
We examine how to scale up text-processing applications that are expressed in a language, Markov Logic, that allows one to express both logical and statistical rules. Our idea is to exploit the observation that to build text-processing applications one must solve a host of common subtasks, e.g., named-entity extraction, relationship discovery, coreference resolution. For some subtasks, there are specialized algorithms that achieve both high quality and high performance. But current general-purpose statistical inference approaches are oblivious to these subtasks and so use a single algorithm independent of the subtasks that they are performing. The result is that general purpose approaches have either lower quality, performance, or both compared to the specialized approaches. To combat this, we present Felix. In Felix programs are expressed in Markov Logic but are executed using a handful of predened