Research And Implementation Of LSTM-based Method For Identifying Explanatory Elements

In product comments,people usually judge on one specific aspect of a product and provide some explanatory comments to explain and support their judgement,then help other customers and merchants to know better of current product.So,explanatory opinion element is the most important part of a commentary.The recognition of explanatory opinion elements also becomes one of the urgent problems that need to be solved,in NLP area.This paper targets Chinese product reviews,using the LSTM framework,studies the detection problem of products' attributes,evaluations and explanatory expression.In particularly,this paper can be drilled down into below 3 problems:(1)Build corpus for explanatory opinion elements:in this paper,we have built the explanatory opinion elements corpus for 2 areas:phone and hotel.And based on the analysis on the commentary text,we gave the definition of explanatory opinion elements,and studied more on the number,distribution of aspect,expression and explanatory expression in different areas.(2)Implementation of LSTM based model:in this paper,we leverage a LSTM based model to recognize the explanatory opinion elements.To get more input information,we also integrated the convolution method.The LSTM's output will be feed into CRF to decode and then get the final label sequence.(3)Integrated system of explanatory opinion elements recognition:in this paper,based on LSTM method,we trained a model which is better fit the explanatory opinion elements recognition tasks,and applied this model in 2 different areas:mobile and hotel.At the same time,we built up an integrated system of explanatory opinion elements recognition and tested with this system.The experiment results show that,the model we are using can do a better job in explanatory opinion elements recognition.