Personalized Recommendation Technology Combined With Natural Language Processing

With the arrival of the Internet information age,information is growing explosively,More and more people are overwhelmed by the huge amount of information on the Internet.The research and development of personalized recommendation system has to be put on the agenda to alleviate the trouble caused by information explosion in the age of big data,This issue is particularly important in the field of e-commerce.An excellent recommendation system can not only help users to choose products better,but also help e-commerce to obtain more profits.At present,recommendation algorithm based on collaborative filtering is widely used in this field,which has achieved good performance to some extent.However,it is still based on collaborative filtering recommendation algorithm to jump out the attention of goods of similar goods,so this paper puts forward a recommendation system combined with deep learning algorithm,can undertake interdisciplinary recommendations,combined with some to some technology in natural language processing,the user purchase of goods that a similar statement in natural language processing.Since the application of deep learning can find some unknown rules in human language through training,it can be migrated to the recommendation,and find some non-intuitive rules when people buy products through deep learning,and apply this rule to the recommendation system.In this paper,we first introduced Item2 Vec technology,which can transform a commodity from an ordinary commodity number to a vector with knowledge representation function through the user's historical behavior data.Similar to the word coding in natural language processing,this technology can effectively improve the effect of model training in the later stage.Due to the complexity of human beings,different people have different values,so the rules in the field of commodity purchase are not universal.In order to make more accurate recommendations,users are clustered and users with similar purchasing behaviors are recommended by using the same trained model.However,due to the high sparsity of e-commerce data,the traditional clustering similarity algorithm cannot work well.Therefore,a new similarity measure is proposed in this paper,which can solve the clustering problem in the case of high sparse data.Then,the commodity recommendation system was constructed by using the sequence to sequence model.The user's behavior record(browsing,collecting,purchasing,etc.)for a period of time is used as input,and the output of model feedback is used as recommendation list.This idea has a good performance in experiments,and I believe it will provide more intelligent and accurate recommendations for human beings in the near future.