Personalization Recommendation Of Dress Combining Knowledge Graph And Emotion Score

This paper studies consumers’ preference for dress attributes and their intention to buy goods.Firstly,the dress of knowledge graph is constructed based on literature review,expert interviews and the research results of dress attributes on e-commerce platforms.Then,through the emotional analysis of online comments on dresses on e-commerce platforms,it constructed a comprehensive product emotional rating model.Then the dress personalized recommendation system is established by fusing the dress knowledge graph and the emotion analysis model.Finally,the accuracy and effectiveness of the system is verified by a comparative experiment on the purchase intention of 20 experienced online shoppers.The paper draws the following conclusions:1.Classify clothing attributes from three dimensions of basic,performance and external,and get 15 clothing attributes,which form a triad of < clothing,attribute and attribute value >,and construct dress knowledge graph through Neo4 j graph database.2.Add 343 clothing stop words to the segmentation of dress online comment text,extract the top 50 high-frequency words,and conclude 5 dress comment feature words.3.Add the dress emotion corpus to SnowNLP database.The experimental verification shows that the average accuracy of the dress emotion rating of SnowNLP database is 82.86% after the expansion of the language data.4.The weight of five comment feature words was obtained through analytic hierarchy process,and the emotional rating model of dress comment was constructed with appearance,price,quality,logistics and service as five indicator layers.5.A personalized dress recommendation system was constructed with knowledge graph and emotion rating model.The comparative experiment verified that the accuracy of the system’s recommendation results was higher than that of the e-commerce platform’s recommendation results,and Kendall’s Tau-C coefficient of the ranking of recommendation results and the ranking of the score list of experimental subjects is about 85.19%,the that indicating the system recommended the sorting accuracy is good.The overall coverage rate of dress types reached 81.2%.