| Networking platforms such as e-commerce website and forum can effectively narrow the distance between manufacturing enterprises and consumers.Enterprises can directly know users experience through on-line reviews and find out products’ quality problem exposed in the using stage,which provides direction for product quality improvement.The article focuses on the online reviews and studied the method of quality level assessment and quality problem mining by quality feature extraction,sentiment analysis and sequential pattern mining.The main contents of this paper are as follows:(1)Extracting quality feature words and constructing semantic tree.First,we see the extraction of quality feature words as the sequence labeling problem,and use CRF algorithm to train the model.We propose six features to train the model that are word,part of speech,content,dependence,governing word and emotional polarity.Semantic tree has been built to describe the semantic relations between quality feature words and we proposed two methods for semantic tree extension: one is the addition of synonymous nodes based on the similarity of words,the other is the addition of subordinate nodes based the correlation of words.(2)Evaluating quality and finding quality problem.We establish a way of computing quality feature words’ emotion score based on the emotional dictionary and analyzes the influence of negative words and degree adverbs on the score.Then,a method for evaluating the quality of feature words is proposed based on the semantic tree and emotional score.We studied the sequential pattern of quality negative sentences and proposed the method to find out quality problem based on sequential pattern mining.(3)Developing product quality analysis system.The system consists of six modules: data fetch,data preprocessing,quality feature word extraction,semantic tree construction,quality evaluation and quality problem finding. |
PDF Full Text Request