Research On Emotional Classification Technology Of Cross - Domain Short Text In Micro - Precision Precision Marketing Platform

Nowadays, with the rapid development of the Internet, social media gradually take people’s lives, microblog marketing has become a newer, more effective marketing tool. And in the development of precision micro-blog marketing strategy, intelligent selection to ordinary Internet users as the main marketing content propagation member of microblog texts and comments on sentiment classification conquest tendentious essential, but in microblogging we encounter the following two problems:First, microblogging text is short of words in 140 words, the feature space sparse and lack of standardization, Therefore, if using the traditional sentiment classification method on microblog short text sentiment classification, accuracy is often poor.Further, since the microblog marketing objects from various differences in very different areas, so that the field associated with the various types of marketing microblogging text objects with vastly different feature distribution of the text, so the traditional sentiment classification methods can not to be used in case of cross-domain situation.For the first issue, the topological characteristics of short text classification based on the polymerization of emotion. First computing the multi dimensions of emotional features of correlation, and by solving strong component in the figure associated with high similar emotion feature extraction from unlabeled corpus. The features similar characteristics to supplement the training set, while reducing training spatial dimensions. Finally, experiments show that the method works on a very short text to alleviate the sparsity and dispersion characteristics of the problem.For the second issue, cross-domain sentiment classification transfer model based on the cost of traffic. Trans the cross-domain sentiment classification problem into the dual optimization problem of maximize the spread between the amount of information and minimize the loss of cross-domain feature fitting, and establish a connection graph of network capacity costs, the minimum cost maximum flow frame the double optimization to solve the problem. Give more intuitive mapping relationship between model description field.Finally, we design and realize the precision marketing platform microblog sentiment classification sub-module, use the above improved sentiment classification method for short text and cross-domain two conditions apply to the actual task of the platform, the platform to help marketers more accurately selected content distribution member. Verify their usage in the real value of the project.