Generation Technology Of Review In Specific Domain Of Social Network Based On LSTM

With the rapid development of Internet technology,online platforms are flooded with reviews from a large number of users.There are even some Internet rumors driven by benefit,which affect the decisions and opinions of netizens about a certain event.In this paper,a text generation model of social network reviews in the specific domain based on long short term memory is proposed to generate reviews.The language description performance of our reviews can be comparable to real social network reviews.The model can be used to provide a large-scale corpus for public opinion guidance.We use the Twitter with complicated application scenarios as the example platform,focus on politics,health,education,entertainment and technology,reviews in which are collected,classified,generated and processed.The main work is as follows.Firstly,review text is classified according to sentence structure,and each review text classifier based on random forest model is designed.Six features are extracted by combining with English grammar,then input into the classifier.The review text is divided into seven types,such as subject-link verb-predicative structure,subject-verb-object structure,imperative sentence structure and so on.Secondly,According to the characteristics of different syntactic structures,LSTM with different parameter integrated with Attention mechanism is designed to learn the styles of different syntactic structures,so as to generate the initial reviews corresponding to the preliminary classification.Thirdly,After observing the initial reviews set,when the text is with significant deviation from the reality,this paper proposes a domain knowledge-based algorithm,which includes text replacement,paraphrase generation and pattern-based customization algorithm,aiming at conducting the bias correction of initial reviews.After that we generate the final reviews that are close to the theme and consistent with the facts.Finally,multidimensional experiments are carried out to verify the effectiveness of the proposed model.In addition,the function of each module is verified,which proves the indispensability and high efficiency of each functional module apparently.The crossplatform adaptability,domain matching degree and repetition rate of the generated reviews are analyzed.