Bootstrapping algorithm for extracting viewpoint evaluation objects based on dependence relationship template

Abstract

The invention relates to a Bootstrapping algorithm for extracting viewpoint evaluation objects based on a dependence relationship template. The grammar and semantic relationships between emotion words and viewpoint evaluation object words are considered, the dependence relationship template between the viewpoint evaluation object words and the emotion words is constructed, and the viewpoint evaluation objects are extracted with the Bootstrapping method. By means of the Bootstrapping algorithm, noise caused by the method that evaluation objects are extracted directly by a vocabulary context is avoided, and the extracting performance of the viewpoint evaluation objects is improved.

Description

technical field [0001] The invention relates to the field of viewpoint mining, in particular to a Bootstrapping algorithm for extracting viewpoint evaluation objects based on dependency relationship templates. Background technique [0002] With the rapid development of the network, a brand-new mode involving a large number of users has emerged on the Internet—Users Generate Content (UGC) mode. On UGC, users publish comments on events, products, and people. These comments are of great value and significance to user purchase decisions, interest mining, personalized information services, public opinion monitoring, and information prediction. However, the number of these comments is huge, the quality is uneven, and most of them are short text descriptions, so manual analysis and mining are extremely difficult. Therefore, the design and implementation of opinion mining tools for comments has become one of the hot issues in the fields of natural language processing and data mining...

AI Technical Summary

Problems solved by technology

Wei Jin et al. marked synonyms and synonyms of candidate opinion evaluation objects as candidate opinion evaluation objects, and then retrained the model, but this method brought many low-frequency words to affect the recognition performance

Shu Zhang et al. used a graph model to identify opinion evaluation objects and emotional words, and regarded the opinion evaluation objects and emotional words as an evaluation collocation relationship pair. Every time m candidate opinion evaluation objects were added, n candidate emotional words were generated, and this was continuously iteratively generated. Opinion evaluation objects and emotional words, the disadvantage of this method is that when evaluating candidate opinion evaluation objects, only the co-occurrence rate of candidate opinion evaluation objects and emotional seed words is considered, and the recall rate is low

Qiu Guang et al. proposed to use small-scale seed emotional words and the dependency relationship between opinion evaluation objects and emotional words to incrementally identify new emotional words and opinion evaluation objects through a two-way communication strategy between opinion evaluation objects and emotional words. However, this method needs to rely on human experience to construct dependency templates

Song Xiaolei et al. used the Bootstrapping method to identify opinion evaluation objects based on the seed opinion evaluation objects, seed word form templates, and part-of-speech templates, but this method ignored the role of emotional words

Images