| Considered as an important way for companies to understand user information and preferences,customer service text content contains a large number of user attributes.How to efficiently extract user attribute information from customer service text has become a hot spot in natural language processing research.Enterprises often pay a lot of attention to some specific attribute tags of users,and will select some user tag content to extract user attributes.However,the current analysis of customer service text mainly focuses on intelligent customer service and user sentiment analysis,besides there are few studies on extracting user attributes based on specified tags.Therefore,this paper extracts the attribute values of tags from customer service text based on the specified user tags,selects machine reading comprehension technology,and makes two aspects:the model improvement technology combined with question keyword recognition and the emotion tag extraction model construction technology that introduces sentence-level pre-training technology.indepth research.First of all,I learned that the machine reading comprehension model always updates the attention while reading the question and text content when reading the question text.Although this method can grasp the question content more comprehensively,the efficiency of reading the question is too low.This paper proposes a question text reading optimization technology based on keyword recognition.First,extract the keywords of the question content to make a question keyword table,and then build the ERNIEK model for extracting the user’s general tag attributes,and identify the key words in the question text when reading the question.The attention is only calculated when the word is used.The improved model has improved the accuracy rate by 2.6%on both the C3 and the dataset used in this paper,which proves the effectiveness of the keyword recognition technology.At the same time,in order to solve the problem that the model only focuses on local text content and ignores the full-text semantics when extracting tag attributes such as user sentiment analysis,this paper proposes a model pre-training technology based on the generated text summary dataset.First,the T5-pegasus model is used to extract customer service text summary data set,and then build a text summary model,train the model to perform text summary extraction tasks to obtain the ability to understand the semantics of the full text.Based on the abstract generation model encoder,the ERNIEA model,a user emotion tag extraction model,was built.The accuracy of this model was increased by 3.8%and 3.6%on the C3 data set and the customer service text data set,respectively.At the same time,the test results on the user emotion label data set also prove the effectiveness of the ERNIEA model.Finally,this paper designs and implements a user tag attribute extraction system based on the MVC framework.It designs a user tag attribute extraction module,a data storage module,and a system display module,and applies the above two tag attribute extraction models proposed in this paper to this system.This system extracts tag attribute values of the input customer service text,stores the operation records and displays the tag extraction results,which can further optimize the service recommendation ability for enterprises. |
