| With the rapid development of Internet technology,there has been a large amount of UGC data on the Internet,such as blogs,forums and so on.How to quickly screen out valuable content from massive information has become a concern in the computer field.Comparative sentence identification and relation extraction are generated in this context.The main task is to identify the comparative sentences in the text and extract the relationship of the comparative sentences.Traditional methods are mostly based on comparative templates and feature engine--ering,which cannot fully make use of the deep semantic information of text.Therefore,this paper proposes an algorithm based on long short-term memory network to solve the problem.The main work of this paper is as follow:(1)This paper collects review data on JingDong Mobile which covers 4 brands such as Xiaomi and Huawei.After removing non-propositional sentences,clauses,and word segmentation,the data is pre-processed to build a corpus for experiments.(2)For the characteristics of comparative sentence identification task,this paper proposes a model based on LSTM and Attention Model.This model can guarantee the distribution of attention probability of input sequence to output sequence and learn the deep semantic features of sentence.Experimental results show that the proposed method is effective and feasible.(3)Aiming at the characteristics of relation extraction task,this paper proposes a model combining BiLSTM and ontology library.The algorithm firstly uses BiLSTM to extract the comparison subject,the comparison object and the comparison element,and then uses the ontology library to mine comparison opinion.Finally,the comparison relation is obtained.The results show that the effect proposed in this paper is effective,but it still needs to be improved. |
