Odinson: A Fast Rule-based Information Extraction Framework

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2183–2191 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2183 Odinson: A Fast Rule-based Information Extraction Framework Marco A. Valenzuela-Escárcega, Gus Hahn-Powell, Dane Bell LUM.AI, Tucson, Arizona, USA {marco,ghp,dane}@lum.ai Abstract We present Odinson, a rule-based information extraction framework, which couples a simple yet powerful pattern language that can operate over multiple representations of text, with a runtime system that operates in near real time. In the Odinson query language, a single pattern may combine regular expressions over surface tokens with regular expressions over graphs such as syntactic dependencies. To guarantee the rapid matching of these patterns, our framework indexes most of the necessary information for matching patterns, including directed graphs such as syntactic dependencies, into a custom Lucene index. Indexing minimizes the amount of expensive pattern matching that must take place at runtime. As a result, the runtime system matches a syntax-based graph traversal in 2.8 seconds in a corpus of over 134 million sentences, nearly 150,000 times faster than its predecessor. Keywords: information extraction, text mining, rule-based methods, information retrieval 1. Introduction Information extraction (IE) forms the backbone of many important real-world applications, such as identifying frag- ments of protein signaling pathways from biomedical pub- lications (Kim et al., 2013) and assembling complex models of real-world systems (Sharp et al., 2019). Efficient meth- ods for scalable retrieval and extraction enable new forms of linguistic analysis leveraging large text corpora (Biber, 2009). At a high level, IE approaches can be grouped into two categories: (a) rule-based methods and (b) machine learning. Chiticariu et al. (2013) found that, while machine learning (ML) approaches have garnered great attention in academia, the opposite is true in industry, where rule-based approaches are preferred because of their interpretability and maintainability (Sculley et al., 2014). Our work fo- cuses on the latter class of approaches. Rule-based approaches make use of patterns that match over a variety of representations, from surface (de Does et al., 2017) to more complex syntactic and semantic struc- tures (Chang and Manning, 2014). While matching patterns over structures like syntax can add resilience to linguistic variation, doing it naively can be very expensive. We argue that this is an important limitation for several reasons. First, slow runtimes limit the applicability for large doc- ument collections. For example, PubMed1, a repository of biomedical publications, has indexed more than one million publications each year, for the past seven years. Sequen- tially analyzing just one year’s worth of these publications with an IE system that takes one minute per paper would re- quire nearly two years. Second, it would be ideal for these systems to be interactive, especially in the rule development stage, and previous research has shown that a slowdown of even a few hundred milliseconds affects the way users in- teract with a system (Brutlag, 2009; Liu and Heer, 2014). Finally, efficiency is critical when incorporating IE systems as components of larger, iterative systems, as …

Content truncated to 500 words