Knowledge-based User Preference Removal And Its Application In Product Personalized Recommendation

The research of natural language understanding is currently one of the hotspots in the field of artificial intelligence,and technological breakthroughs based on this as the core and the implementation of related models are also emerging one after another.At present,the application of deep learning in the customization of industrial products has not yet been promoted.With the rise of intelligent manufacturing and the advancement of Internet technology,in the future,the use of natural language understanding to achieve personalized recommendations for users' products will inevitably usher in a larger market space.To provide users with better personalized recommendations,the key is to accurately understand the user's preference characteristics.When the computer understands the user's preferences,the first thing to solve is the ambiguity problem in natural language understanding.This paper analyzes the current research status of natural language understanding,word sense disambiguation,and recommendation systems at home and abroad,and proposes a method to transform word sense disambiguation tasks into deep learning text classification tasks.This method makes full use of the advantages of the deep learning model to learn the basic relationships and language features between the corpus,avoids the problems that arise in the analysis of various complex sentence components and relationships,and uses specific data sets to verify the correctness of the method.This article first summarizes and analyzes various sentence-level and word-level ambiguities in the process of natural language understanding,and focuses on the phenomenon of word sense disambiguation,and demonstrates in detail the auxiliary role of knowledge in word sense disambiguation.Secondly,it is proposed to use the BERT model to complete the task of word sense disambiguation.In order to verify the feasibility of the model and illustrate the auxiliary role of knowledge,this paper designs the composition structure of auxiliary knowledge "domain+ attribute" in word sense disambiguation,and enhances the language representation of the model.ability.Based on the distinct characteristics of words having "one meaning item,one domain",the concept dependency tree is established,and the concept of the concept dependency tree of the word to be disambiguated is gradually refined from the root node to the leaf node,and the domain of the disambiguation word is located to narrow it down.The category of knowledge.This article uses crawler tools to obtain encyclopedic knowledge,applies it to the mechanical field data set established by itself,and uses the sorted data as the attribute information of auxiliary knowledge.The words to be disambiguated in the sentence and the auxiliary knowledge form the form of word sense pairs,and the disambiguation task is transformed into a classification task of whether the sense items can explain the meaning of the words to be disambiguated in the sentence.Next,for the problem of randomly masking Chinese characters in the MLM task of the BERT model but ignoring the relationship between Chinese characters,this paper replaces the "covering" based on Chinese characters in the model with the word-based "covering".Through the verification of the data set,the effect of the model is improved by two percentage point.Finally,this paper proposes an industrial product personalized recommendation prototype system that integrates word sense disambiguation,designs the prototype system structure,and discusses the implementation process of each component module.The specific dialogue example is applied to the prototype system to realize the disambiguation of user needs first,and then extract the preferences.Finally,a good recommendation result is obtained,which verifies the feasibility of the model.