Semantic Dependency Analysis For Patent Text

Semantic Dependency Analysis is an effective semantic analysis method. Both of thehierarchical structure information and the semantic relations between elements are veryimportant in syntactic analysis. The treebank annotation standards in this paper can expressboth the type information of chunk and semantic information. Semantic Dependency Analysiswith the chunk as a unit can take into account hierarchical structure,dependency structure andsemantic information at the same time. This paper explores the technology of semanticdependency analysis by combining rules and statistical methods. The following sectionsdescribes the main research content.First, Building Chinese semantic tree bank of patent text. On the basis of3DimensionalDynamic Concept Model, we determined the label set that makes chunk as annotation object.This paper proposed top-down building process and determined markup format of tree bank intext. And we defined tree bank form that has level for sentence and can reflect the core chunk.The work is completed by human with the help of computer.Second, recognition of different semantic chunks.The most outer chunks is consistentwith the normal sequential of subject-predicate structure. But some verbs are nominalized inthe inner chunks, which makes the sequential change. According to the structure features of asentence, this paper proposed top-down recognition method of different semantic chunks. Itincludes the recognition of outer chunks and the recognition of inner chunks. The F value ofthe outer chunks and inner chunks are83.36%and79.68%respectively. The overall F value is82.2%.Third, semantic dependency analysis by utilizing the rule-based method and thestatistical method. The process includes determining the dependency arcs and judging therelationship. It can make sentence structure become more simple based on chunk as unit. Sothis paper determine the dependency arcs by rules. On the basis of correct chunks recognition,the accuracy rate is96%with rules. Then judge the semantic relationship based on CascadedConditional Random Fields and modify the relationship with rules. The F value of outer semantic relationship is80.96%and the F value of inner semantic relationship is76.75%.The method of building treebank is researched in this paper.This paper determines thelabel set,markup format of tree bank in text and tree bank form. This paper proposedtop-down building process. On the basis, proposed top-down recognition method of semanticchunks. By utilizing the advantage of the rule-based method and the statistical method, thispaper researches on the semantic dependency analysis. The F value is73.94%.