【目的/意義】準確挖掘消費者在線評論情感傾向,對于改善商家服務具有重要意義,而情感傾向預測的準確性仍需提高。【方法/過程】文中設計基于集成學習的在線評論情感分類算法,即以N-gram算法分析在線評論詞語特征,結合情感詞典構造文本特征,利用邏輯回歸、Light GBM等機器學習方法為基礎的集成學習進行訓練,實現在線評論情感分類。【結果/結論】實現了評論的情感傾向預測,在電腦評論數據集,較之于經典的SVM算法和無監督類算法,該模型的分類衡量指標F1值分別提高了10%到30%不等。同時,在酒店、圖書等不同領域的數據集上顯示,該方法的分類準確性仍具有上述效果,證明了該方法具有領域移植性。

【Purpose/significance】It is of great significance for business services improvement to explore online emotional tendencies of consumers accurately, while the accuracy of emotional tendencies prediction is still needed to be enhanced.【Method/process】In this paper, an online emotion reviews classification algorithm based on integrated learning is designed which uses N-gram algorithm to analyze the characteristics of online reviews words. Text features are constructed with emotional dictionary,integrated studying & training are carried out with machine learning method including logistic regression and Light GBM to realize online emotion reviews classification.【Result/conclusion】The emotional tendency forecast for reviews has been realized. In computer data reviews area, the classification indicators F1 for this module have been improved10% to 30% or so compared with classical SVM and unsupervised algorithm. Meanwhile, there is same effect for the classification accuracy for this method in data displaying area of hotels and libraries. It has been proved that this method has domain transplantability.