Fine-Grained Type-Activated Hierarchical Analytic Engine
Publication Date: 2014-Apr-22
The IP.com Prior Art Database
A method and system is disclosed for dynamically re-ordering execution of annotators based on types of annotators produced by each stage of analysis.
Page 01 of 3
Disclosed is a method and system for dynamically re-ordering execution of annotators based on types of annotators produced by each stage of analysis.
Traditional unstructured information analytics known as annotators process information in a very procedural way. For example, the unstructured information analytics have a set of word lists for recognizing words in text and there are processing rules that always check preconditions of the unstructured information analytics before firing. Information is gathered into a central storage location and each analytic that participates in the analysis resides in a pipeline configuration. This potentially results in a large amount of
wasted processing because each analytic in the pipeline must evaluate the document and central storage system to determine whether to emit additional annotations. Many times, an analytic is only executed to analyze certain aspects of a document resulting in either additional burden to maintain multiple configurations for a monolithic annotator or large amounts of wasted work during processing time.
Most systems in the current state of the art attempt to analyze unstructured content at the granularity of a so called document in a sort of top down approach where each stage of the processing is presented with a semi structured form of the document. The top down approach does not apply deeper knowledge of types of annotators to the system. This is a coarser approach to the problem and is effectively inefficient in terms of processing and resources consumed at each stage of analysis.
The disclosed method and system utilizes a rule type hierarchy for triggering the re-ordering. The method and system utilizes one or more word lists as initial triggers
with the processing following a functional model and determines the types of the annotators from at least one of the one or more word lists and a processing rule. After determining the types of annotators, the method and system en-queues all rules that depend on an output type for execution.
Additionally, the disclosed method and system utilizes one or more object oriented principles used for allowing one or more hierarchical word sets and one or more rule sets. If a word/rule hierarchy were envisioned as a tree structure, the method and system exhibits an "is-a" relationship with words and rules above it. Therefore, the method and system executes all rules at each node of the tree above the word when a
word in the terminal node was encountered. If a word were encountered in the middle of the tree, the method and system executes only the rules associated with the one or more word lists and above that point in the tree.
For the dynamic re-ordering, the disclosed method and system firstly activates and runs the annotators that have no input dependencies....