Research On Aspect-based Sentiment Analysis Based On Attentional Interactive Network

Aspect-based Sentiment Analysis is a research hotspot in sentiment analysis in re-cent years.It can extract sentiment information of specific aspects in texts,and it has more application value than coarse-grained sentiment analysis.Studies have confirmed that the semantic attention interaction between context and aspect words is very important for this task,but the existing methods are not sufficient for the exploration of the interac-tion information between the two,such as:(1)Semantic fusion is not considered in the attention interaction process,however accurately judging the sentiment corresponding to aspect words not only depends on the context,but also needs to combine the semantics of the two?(2)The extraction of sentiment features is insufficient and inaccurate,including that the representation of aspect words does not fully reflect the semantics of the context,and non-feature-level calculations is not accurate enough?(3)Mainly focus on the absolute position of words,rather than the relative position which has more semantic information.In response to the above problems,this paper proposes a Attentional Interactive Fu-sion Network for Aspect-based Sentiment Analysis,which uses functional relative po-sition encoding to introduce position information,through a semantic fusion operation combined with a multi-head attention mechanism and a multi-dimensional attention mech-anism to achieve semantic fusion and interaction between context and aspect words,the experimental results show that the accuracy and Macro-F1 of AIFN on the two mainstream datasets are superior to the current advanced models.In addition,existing methods only focus on the semantic connection between aspect words and context,but do not pay attention to the interrelationship and dependency in-formation between aspect words.In response to this problem,on the basis of AIFN,this paper proposes an Attentional Interactive Fusion Network with Aspect Dependencies for Aspect-based Sentiment Analysis,which uses the simplified AIFN as the sentiment feature extractor of aspect words,and uses multiple attention mechanisms to capture the depen-dency information between aspect words.The experimental results show that AD-AIFN can learn the dependency information between aspects,and these information can further improve the performance of this task.Analyzing customer evaluations occupies a considerable proportion in the practical application of sentiment analysis.With regards to this,this paper designs and implements a general and open aspect-based sentiment analysis system for customer evaluations.In addition to providing aspect-based sentiment analysis for various evaluations,various re-ports and visual display functions,the system also has model training functions to meet the personalized needs of users.At present,the system already contains 7 sentiment analysis models for users to choose from,including the models proposed in this paper and other 5advanced models.Due to its open architecture design,new models can be added easily in the future.The goal of the system is not only to provide a shared utility tool for ordinary users,but also to help researchers verify and compare various sentiment analysis models.