Research On Opinion Mining Of Product Reviews Based On Dependency Syntactic Rules

With the rapid development of e-commerce,"online shopping" has become a new consumption mode.The comment function of each online shopping platform encourages people to evaluate their use feelings after consumption.Tens of thousands of reviews data contain different views of consumers on different aspects of products.Due to the different sentence structure and expression of each comment,it is very timeconsuming to process and analyze it manually.In the face of such a problem,opinion mining technology for comment text was born.Opinion mining of product reviews is an extremely important decision-making basis for both consumers and businesses: for consumers,mining the opinion information in reviews can provide reference during purchase,timely grasp the reputation and advantages and disadvantages of products,and make the best purchase decision;For businesses,through view mining technology,we can intuitively and quickly understand consumers’ view tendency on a product or even specific product characteristics,and timely grasp the market dynamics,so as to make targeted optimization decisions and improve market competitiveness.At present,the research on opinion mining of reviews is roughly divided into supervised methods based on machine learning and unsupervised methods based on syntactic dictionary.Although the machine learning method is outstanding,the model does not fit well for sentences with complex structure;The unsupervised method mainly relies on the grammatical and semantic collocation between opinion elements,which is more universal and interpretative.Nowadays,most unsupervised opinion mining adopts dependency parsing,which extracts and analyzes text fragments related to opinions from reviews by constructing syntactic rules or templates.In view of the problems existing in the current research of extracting views through dependent syntactic rules,this paper proposes a method of establishing 17 dependent syntactic rules from the three perspectives of evaluation object,emotional word and core predicate to extract the binary of opinion collocation of < evaluation object,emotional word >,so as to increase the accuracy and coverage of opinion mining.In order to expand the semantic boundary between the evaluation object and emotional words,this paper proposes two post order syntactic rules,ATT chain and COO chain,after 17 pre order rules,so as to make the opinion semantics more complete.The experimental results show good performance in the values of P,R and F1,which effectively proves the feasibility of this method and its adaptability in product reviews in different fields.In addition,in view of the incomplete extraction of adverbs in opinion mining,after extracting the opinion collocation binary,this paper proposes 4adv syntactic rules for different situations to extract degree adverbs or negatives,form an opinion collocation triplet of < evaluation object,emotion word,degree word >,and expand the opinion mining work.In order to integrate and analyze the extracted opinions,this paper constructs product feature dictionaries in different fields through word2 vec technology to form a more complete opinion collocation quad of < product features,specific feature description,emotion words,degree words >.Finally,according to the emotion quantification formula,calculate the emotional tendency value for each opinion,integrate the opinion scores belonging to the same product features and visualize them in the form of charts,so as to provide decision-making basis for consumers to buy products and merchants to improve products.