Research And Implementation Of User Suggestion Mining In Product Reviews

With the rapid development of the Internet and e-commerce,more and more people shop online.In the case of fierce market competition,merchants need to find out some new consumer demand for products in time to improve market acceptance and competitiveness.Product reviews include some specific suggestions given by users to the product or the merchant.The company can improve the follow-up products and formulate corresponding marketing strategies according to these suggestions.It has very important application value and needs to be explored.In the user suggestion mining study in product reviews,the first task is to detect comments containing user suggestions from a large number of product reviews.This thesis studies and solves some problems of user suggestion mining in product reviews,such as lack of relevant data sets,flexible Chinese expressions,difficult feature construction,and large manual workload.Based on this,the main module of user suggestion mining prototype system in product reviews is built.Specifically,the main work of this thesis is as follows:1.For the problem of lack of annotated dataset,crawling a number of reviews and constructing a Chinese suggestion mining dataset for product reviews manually.Then an ensemble learning model is proposed to detect the suggestion sentences.The model firstly uses Stacking to combine classifiers and constructed probabilistic feature space.Then,the convolution neural network(CNN)and paragraph vector(PV)model are used to construct the CNN feature space and paragraph vector feature space of the reviews respectively.Finally,the above features are fused and the Bagging classifier was trained to classify suggestion sentences.Experimental results on Chinese dataset verify the effectiveness of the model.2.Considering the problems of complex rules,large labeling workload,high feature dimension and sparse data in the traditional suggestion sentences classification method,an unsupervised suggestion sentences classification method based on PU learning is proposed.Firstly,some suggestion sentences are selected from an unlabeled review set by using a simple rule to form a positive example set.Then a reliable negative example set is constructed by Spy technique in the feature space of AutoEncoder neural network to reduce feature dimensions and alleviate data sparsity.Finally,Multi-Layer Perceptron(MLP)is trained on the positive example set and the reliable negative example set to classify the remaining unlabeled samples.The experimental results show that the proposed method can classify suggestion sentences effectively without manually labelling the data.3.The main functional modules in the user suggestion mining prototype system in the product reviews are built,including data collection,data preprocessing,suggestion statement detection and suggestion information extraction module,which can automatically detect the suggestion statement and extract the suggestion information from the product review.