Research And Implementation Of Chinese Explanatory Opinion Extraction Method Based On Relationship Recognition

With the popularity of the mobile Internet and the arrival of the artificial intelligence wave,a lot of data processing is increasingly becoming a means of achieving technological breakthroughs and commercial applications.With the popularity of commercial platforms,comments are often more convincing than content.The task of opinion relation extraction is to present the unstructured information of opinion elements to users in the form of structured tuples.Entity co-resolution task is the process of translating different representations of the same entity into equivalence classes.Compared with the traditional relational extraction task,the extraction result will not be further subdivided according to entity differences,for explanatory product data,the opinion tuple results are divided according to the entity’s equivalence class,which can often help users or merchants make decisions.Therefore,this paper combines the recognition of opinion relationship with the co-digestion of opinion entity.The content is mainly condensed into three points,namely,the Identification of Interpretative Opinion Relation,the Opinion Entity Co-Reference Digestion and the realization of interpretative opinion information extraction system.Interpretive opinion relation recognition task.For the problem of missing context information for opinion elements,this paper first on the basis of relationship identification,the frame of opinion relation extraction based on target attention and the frame of opinion relation extraction integrated with Transformer coding end are proposed,explore the performance of opinion relationship extraction and extract opinion tuples.Then,for the complex cross-cutting relationships in the comments,we further adopted the pre-trained Elmo network layer strategy and pre-trained Word-Embedding strategy,improve the semantic representation of opinion elements in sentences,exploring the performance of opinion relationship extraction tasks under different pre-training strategies.The Opinion Entity Co-Reference Digestion task.In order to eliminate the cofingering phenomenon existing in the opinion relationship,that is,to eliminate that a kind of entity object is expressed by multiple entities.This paper propose a target-based attention-based entity co-finding digestion framework and Target Attentional Framework with fusion Transformer Encoding,through the selection of multiple features,take the output of the opinion relationship extraction task as input,eliminate entity ambiguity in opinion tuples.Explanatory opinion relationship extraction system implementation.The content of this paper is presented through the opinion relation extraction system.First,we extract the information in the explanatory corpus in the form of tuples.Based on this,we will extract the tuple classification for inter-entity co-finger digestion.Finally,the results are presented to the user in the form of a tuple equivalent.