Web series, which is broadcasted on the network, has been developing rapidly through the advancement of mobile network and electronic commerce. This paper aims to predict video views of web series based on comment sen...
详细信息
Web series, which is broadcasted on the network, has been developing rapidly through the advancement of mobile network and electronic commerce. This paper aims to predict video views of web series based on comment sentiment analysis and improved stacking ensemble model. Apart from conventional variables, sentiment score variables calculated from viewer comments were added as input variables. Based on sentiment lexicons built with smooth so-pmi algorithm, we calculated sentiment scores of comments by assigning weights to modifiers and the number of "likes ". We proposed the improved stacking ensemble model for prediction, which utilizes the precision weighted average method. Random Forest, Gradient Boosting Decision Tree, Extreme Gradient Boosting and Light Gradient Boosting Machine were taken as base learners of the stacking model. The results showed that by adding sentiment score variables, the improved stacking ensemble model can further improve the predictive performances.
Mining based on opinions can extract useful information from users' comments. After doing cluster and analysis on the information, users can get a detailed understanding of the commodity, then determine to buy the...
详细信息
ISBN:
(纸本)9781467382700
Mining based on opinions can extract useful information from users' comments. After doing cluster and analysis on the information, users can get a detailed understanding of the commodity, then determine to buy the commodity or not. In this paper, firstly, we extract evaluation objects and evaluation words, then cluster the evaluation objects. Next based on so-pmi algorithm, judge the polarity of evaluation words and determine their polarity intensity values, then use K-means clustering algorithm to cluster the evaluation words. Last, for every kind of target evaluation object, make a count on the proportion of each kind evaluation word, and show the result to users in an intuitive way. This paper uses noun phrase pattern to match comments to extract evaluation objects and put forward the thematic words extraction algorithm. On judging the evaluation words' polarity, this paper establishes an emotional seed dictionary for each target object. The method of establishing dictionary for every attribute can eliminate the influence that less-correlation evaluation words have on the polarity judgment.
暂无评论