ABSTRACT
In recent years, BBS, blog, website of electronic business affairs, network
application have been widely popularized with the rapid development
of information technology, which has produced a large number subjective text by the
user, such as people’s review information on electronic commerce website product.
These reviews contain subjective feelings of customer behind buy and the use of
products. If we summarize and analysis the user reviews, it makes us know how many
users hold a positive attitude to the product, how many users hold a negative attitude.
This will help us to understand the user emotional attitude of the
commodity, for potential users and businesses make a decision. Therefore, digging
out valuable information from the mass of product reviews is very important. It’s time-
consuming, cumbersome and low efficiency to obtain people’s tendency on a topic (or
product) of the emotional from the mass information by artificial reading way. So, how
to use the computer to automatically determine the user sentiment tendency based
on text emotion analyzing is a research topic that with good value of application and
popularization.
In this paper, we can find that when analyzing emotional tendency, emotion
clustering is an important step, clustering results will affect the judgment of
emotional tendency based on the research on analysis of the emotion. But the currently
traditional clustering algorithms that are used have a low accuracy of clustering results.
Therefore, this paper proposes an emotion analyzing method based on spectral
clustering. Compared with the traditional clustering algorithms, it is simple, and its
performance is better than the traditional clustering algorithm.
The content of this paper is mainly divided into the following several aspects:
First of all, this paper introduces and analyses the traditional text sentiment
analysis technology amply. Through the analysis of the performance
of traditional algorithm, we find that when make sentiment analysis on text by
traditional methods, it’s often confined using of phrases and syntactic information to
identify the opinion of words and phrases, which ignore the influence
of the context words and intention of user. This makes a inaccuracy of the text
clustering results, affecting the text sentiment orientate on judgment.
Secondly, this paper put forward the improved emotion analysis method based
on the spectral clustering. The method obtains the feature vector by the decomposition
of Laplace matrix firstly. These feature vectors corresponding to different feature
information of data. Then user reviews a part of the feature information to determine the
direction of clustering. Finally, the system makes clustering automatically according to
user's selection, thus obtained emotional classification that user needs, which improves
a certain degree of accuracy on clustering result.
Finally, the improved algorithm is applied to the mining of online product
reviews opinion. The system is designed mainly based on the improved algorithm
proposed in this paper. Description of the design of the key part of the model is in