Research On Aspect Level Affective Tendency Analysis Of Comment Text Based On Deep Learning

With As society progresses,more and more consumers are using review information from e-commerce platforms as an important reference indicator for product selection,and massive amounts of review information are increasingly favored by users.People are more inclined to pay attention to product review information to complete their purchasing decisions.Analyzing review text data is of great benefit to both consumers and platforms,and it also promotes the development of aspect-level sentiment disposition analysis.Aspect-level sentiment analysis is an important task in sentiment analysis.The aim is to determine the sentiment propensity of multiple attribute facets,i.e.,multiple aspects,of an entity in a review text by learning the semantic information of the text.In this paper,an in-depth analysis of existing aspect-level sentiment classification algorithms based on attention mechanisms reveals the following problems.First,the existing attention mechanisms do not combine the context of the whole review text in the weight assignment of sentiment words and aspect words.In particular,the direct use of deep learning models to analyze the full text leads to an unbalanced weight distribution of aspect words and sentiment words.Second,the existing algorithms not only lack the utilization of prior knowledge,but also the phenomenon of multiple meanings of sentiment words has not been addressed.To address the above problems,the main research of this paper is as follows.(1)The sentiment analysis algorithm combining sentiment lexicon and multiheaded attention mechanism is proposed for the extraction of aspect words and sentiment words.Firstly,it extracts entity and relationship of aspect words and sentiment words based on syntactic dependencies to obtain the location information of both,and expands the existing traditional sentiment lexicon,and sentiment words cover the sentiment tendency expressed in different contexts;secondly,it unifies the modeling of aspect words,sentiment words and contexts by multi-headed attention mechanism,calculates the feature weights of aspect words and sentiment words in contexts,and forms a feature matrix;finally,it is input to convolutional neural network to realize the interaction between the two.Finally,it is input to the convolutional neural network to realize the interaction between the two and complete the aspect-level sentiment tendency classification task.(2)In order to solve the problem of semantic joint modeling of evaluation aspect words and context,a strong correlation between evaluation aspect words and contextual text is considered in the research process.The interactive attention mechanism combined with location-aware aspect-level sentiment propensity analysis algorithm is proposed,and the interactive attention mechanism consists of two parts which are global attention mechanism and local attention mechanism.The global attention mechanism extracts the features of the whole text,and the local attention mechanism adjusts the weights of aspect words in the text by combining the given aspect word categories.The relative position information of aspect words and sentiment words is used to perceive the position and form a binary group of aspect words and sentiment words.The algorithm classifies sentiment words in combination with contextual context,and experimental results demonstrate the results of the interactive attention mechanism on the AI Challenge 2018 and web review datasets,increasing the accuracy by 0.79% and 0.93%,respectively,over the suboptimal model.