Research And Application Of Marketing Intention Recognition Based On Ensemble Learning And Topic Memory Network

Due to the rapid development of Internet information and the increasing number of people who love news,the way news is disseminated has changed.Although the existing self-media operators have made tremendous contributions to the problem of safe Internet access,the network is open,and it still cannot completely prevent malicious online marketing events,so it is very important to identify marketing intentions.At present,the main research on marketing intentions is based on external resources for effective identification,but due to the characteristics of external resources that are difficult to obtain,some methods of intention identification have not been widely used.By referring to domestic and foreign literatures,the three main methods of marketing intent recognition are:matching through log information;rule judgment using a dictionary as a template;and intent recognition based on a classification method.Because the corresponding data set is difficult to find,it is difficult to automate the way of log information and dictionary,so the classification method is more superior.Due to the sparse nature of short news text data and the huge size of the data set,it is essential to optimize the feature extraction method and build a classification model with strong generalization ability.In view of the above problems,this paper proposes a marketing intention recognition method based on integrated learning and topic memory network,and does the following work:(1)The LSI-Word2 vec topic extraction method is constructed.Because the Word2 vec algorithm has superiority in the document expression direction,the LSI-Word2 vec model is established,so that the advantages of the algorithm are fully utilized in the feature extraction process.(2)The decision pruning strategy is integrated into the first layer model in Stacking to improve the operation efficiency,and an efficient integration method in machine learning is constructed.By combining the prediction information of multiple models,it has the effect of higher accuracy.(3)Research and analyze the recognition of marketing intention based on the topic memory network.In order to prevent the gradient explosion problem,the topic extractionmethod is optimized,which can extract semantics efficiently and achieve efficient marketing intention recognition effect.(4)The CLSTM classifier is built.For the classifier CNN in TMN,it can extract local information and cannot capture remote dependencies.After the LSTM sequence layer is added,the timing relationship of elements can be added,which can not only correlate the sentences between sentences Local information,and can be dependent on long distances between sentences.(5)Design and implement a set of marketing intention recognition system.Under the real network environment,by collecting news data from media websites and extracting effective text features,the purpose of identifying marketing intentions can be achieved.