Loading...

Table of Content

      
    20 November 2014
    Volume 49 Issue 11
    Sentiment analysis on Chinese Micro-blog corpus
    LUO Yi, LI Li, TAN Song-bo, CHENG Xue-qi
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  1-7.  doi:10.6040/j.issn.1671-9352.3.2014.194
    Abstract ( 834 )   PDF (1457KB) ( 1408 )   Save
    References | Related Articles | Metrics
    The rise and spread of Micro-blog make sentiment classification on short texts become a hot area. A new method was proposed for Micro-blog sentiment classification. First of all, this method will create an emotional dictionary with two-levels, and the words for different levels will get different enhancement; then in order to get features, N-gram method was used, which found new emotional words and emotional information from a short text. The experiment results show this approach has improved precision and recall rate compared to the traditional ways. This algorithm also did a very good job in COAE 2014.
    Micro-blog orientation analysis based on emotion symbol
    LIU Pei-yu, ZHANG Yan-hui, ZHU Zhen-fang, XUN Jing
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  8-13.  doi:10.6040/j.issn.1671-9352.3.2014.051
    Abstract ( 555 )   PDF (741KB) ( 1173 )   Save
    References | Related Articles | Metrics
    At present, the researches of Micro-blog orientation analysis mainly concentrate in the text, without considering the impact of other emotional factors. By analyzing and studying Sina Micro-blog, new words and emoticons dictionary were added into special Micro-blog dictionary with traditional emotional dictionary. Meanwhile, rhetoric and sentence were analyzed in this paper to improve the effect of orientation analysis. The experimental results showed that the method can obtaine better performance in Micro-blog orientation analysis.
    Sentiment classification method of Chinese Micro-blog based on semantic analysis
    YANG Jia-neng, YANG Ai-min, ZHOU Yong-mei
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  14-21.  doi:10.6040/j.issn.1671-9352.3.2014.069
    Abstract ( 875 )   PDF (1943KB) ( 2029 )   Save
    References | Related Articles | Metrics
    By analyzing the structural features of Chinese Micro-blog, a sentiment classification method based on semantic analysis was proposed. For the proposed method, firstly, an emoticons sentiment lexicon and a network language sentiment lexicon were built. Then by using these two lexicons and the dependency parsing results, the sentiment expression Binary Tree was constructed. Finally, the sentiment strength, which was calculated by the established rules, was applied into the sentiment classification. Experimental results show that this method is effective and two built sentiment lexicons can better enhance the performance of the sentiment analysis system.
    Micro-blog opinion analysis based on syntactic dependency and feature combination
    XIA Meng-nan, DU Yong-ping, ZUO Ben-xin
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  22-30.  doi:10.6040/j.issn.1671-9352.3.2014.074
    Abstract ( 671 )   PDF (1954KB) ( 1653 )   Save
    References | Related Articles | Metrics
    Micro-blog opinion mining faces the difficulty because of the short text's conciseness. The technique of syntactic dependency relation analysis and CRFs(Conditional Random Fields) were combined to extract the candidate opinion objects. And then the dictionaries of the opinion analysis and all kinds of semantic features were used in the machine learning method to improve the performance of the opinion classification. The precision, recall and F1 values were used as the evaluation metric. The experimental results on the COAE(Chinese opinion analysis evaluation) data set verify both the validity of emotion factor extraction approach and the impact on opinion classification performance by different features. The macro and micro precisions for the opinion classification task are both 91.4%.
    Personalized ranking of Micro-blogging forwarders
    KUANG Chong, LIU Zhi-yuan, SUN Mao-song
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  31-36.  doi:10.6040/j.issn.1671-9352.3.2014.305
    Abstract ( 683 )   PDF (1411KB) ( 2245 )   Save
    References | Related Articles | Metrics
    The repost action is the main way for information spreading in Micro-blogging platform. Nowadays, many works have been done focusing on the repost behaviors' analysis and prediction. However, the problem about how to find the users who are the most likely to repost a given Micro-blog remains unsolved. In this paper, a general predictor, which combines Bayesian Personalized Ranking optimization criterion with Factorization Machines was presented to predict the reposter of a microblog. Furthermore, factors which affect a user to be a reposter were analyzed in details. With these facts, prediction of the reposters over large-scale real datasets was conducted. The experiment proves that this method can improve the effect of the prediction obviously. Meanwhile, method based on pair-wise and feature-related can solve the prediction problem more efficiently.
    Sentiment analysis of Chinese Micro-blog based on semi-supervised learning
    ZHU Xi, DONG Xi-shuang, GUAN Yi, LIU Zhi-guang
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  37-42.  doi:10.6040/j.issn.1671-9352.3.2014.136
    Abstract ( 611 )   PDF (1003KB) ( 1490 )   Save
    References | Related Articles | Metrics
    Sentiment analysis of Chinese Micro-blog usually refers to classification of Micro-blogs into positive, negative and neutral polarity. According to the characteristics of Micro-blogs, such as fragmentation and imbalanced of sentiment class, on the basis of reserved self-training method we presented before, text features were extracted that are appropriate for the sentiment analysis of Micro-blog, and then a training degree threshold setup method was proposed to optimize the iteration termination condition of reserved self-training method. These methods not only take advantage of the effective treatment on imbalanced distribution problem but also prevent the overtraining problem in training process. The evaluation result in COAE2014 showed the effectiveness of these methods.
    Comparative study of methods for Micro-blog sentiment evaluation tasks
    SUN Song-tao, HE Yan-xiang, CAI Rui, LI Fei, HE Fei-yan
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  43-50.  doi:10.6040/j.issn.1671-9352.3.2014.016
    Abstract ( 459 )   PDF (1260KB) ( 318 )   Save
    References | Related Articles | Metrics
    This paper was a report on COAE2014. The methods to solve the tasks were described, and deeply analyzed by referring to the results. There were 5 different tasks in this year's contest, 3 of which were related to Micro-blog and were focused in this paper. In the new sentiment words discovering and determining of Micro-blog task, the important processes was extracting candidate new words by using the alignment results of Google translation service, then filtering frequent words by ranking their PMI. In the sentiment classification of Micro-blog task, two different methods were used to solve the problem. One was based on sentiment lexicon which was the traditional method. The other was based on CRFs combining the sentiment lexicon. The last task was to extract opinion aspects from Micro-blog and then to determine the sentiment on them. Firstly, the phrases that represent the products' name and aspects were extracted according the betweenness and closeness of the complex network formed by all the nouns in two steps respectively. Then, three methods were introduced to extract the exact product aspects and its sentiment. The first one was based on simple rules which extracted phrases in the sliding window. The other two were supervised learning procedures which were all based on CRFs.
    New methods for extracting emotional words based on distributed representations of words
    YANG Yang, LIU Long-fei, WEI Xian-hui, LIN Hong-fei
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  51-58.  doi:10.6040/j.issn.1671-9352.3.2014.255
    Abstract ( 1096 )   PDF (3225KB) ( 3185 )   Save
    References | Related Articles | Metrics
    Word-level sentiment analysis is a hot research interest in the field of affective computing. How to recognize and analyze these new emotional words automatically becomes an urgent problem. Firstly, statistics-based approach was used to identify the new words in Micro-blog corpus and then distributed representation of new words was trained by using neural network in order to get the correlation between words in corpus. Finally three vector-based methods to find new emotional words were introduced. The experimental results indicate that the proposed methods in this paper can be effectively used in discovery of new emotional words.
    Method of implicit discourse relation detection based on semantics scenario
    YAN Wei-rong, HONG Yu, ZHU Shan-shan, CHE Ting-ting, YAO Jian-min, ZHU Qiao-ming
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  59-67.  doi:10.6040/j.issn.1671-9352.3.2014.077
    Abstract ( 473 )   PDF (1880KB) ( 444 )   Save
    References | Related Articles | Metrics
    The implicit discourse relation detection has a higher difficulty. For this, a method was proposed to detect implicit discourse relation based on semantics scenario. The compression of description form was realized by frame semantics that abstract argument as conceptual semantic description (semantics scenario), and then mine the comparable argument pairs through semantics scenario from large-scale static data. It can ensure accuracy while improve detection efficiency. The discourse relation was detected in Penn Discourse Treebank (PDTB). The accuracy can reach to 55.26%.
    Key sentiment sentence prediction using SVM and RNN
    LIU Ming, ZAN Hong-ying, YUAN Hui-bin
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  68-73.  doi:10.6040/j.issn.1671-9352.3.2014.025
    Abstract ( 969 )   PDF (1423KB) ( 1568 )   Save
    References | Related Articles | Metrics
    Key sentiment sentences play an important role in predicting the sentiment distribution in texts, and therefore it improves the performance after correctly judging these key sentences. After analyzing the advantages and disadvantages of the state-of-the-art approaches which are mainly based on rules and statistics, it is found that rule-based methods achieve high accuracy but with low coverage, the statistic method is quite the opposite. In this paper, a novel deep learning framework to predict sentiment distributions based on Recursive Neural Network as well as Support Vector Machine was introduced. There are sentiment features including not only grammar information such as sentiment and negative words, but also statistical information like word vector in deep learning. Meanwhile, text features like sentence pattern and position were also involved. This method combines SVM and RNN in deep learning to predict sentiment distributions in texts, which outperforms other traditional approaches. The result from COAE2014 Task 1 shows that our method achieves a MicroF1 value of 0.388, higher than the average level.
    Feature selection algorithm based on sentiment topic model
    ZHENG Yan, PANG Lin, BI Hui, LIU Wei, CHENG Gong
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  74-81.  doi:10.6040/j.issn.1671-9352.3.2014.328
    Abstract ( 764 )   PDF (1129KB) ( 1046 )   Save
    References | Related Articles | Metrics
    In order to exert potential commercial value and social value of subjectivity text in enterprise business intelligence and public opinion survey and so on, a novel feature selection algorithm based on sentiment topic model was proposed, which takes both opinion term and opinion co-occurrence term into consideration to help topic modeling, and then the conditional distributions of opinion term in positive topic and negative topic were effectively estimated. This method tries to measure the importance of opinion feature in sentiment orientation. SVM was used in the experimental stage for classification.The experiment result shows that the algorithm has a higher recognition ratio and offers practical capabilities for cross-domain.
    A trusted inter-domain access control scheme for enterprise WLAN
    LÜ Meng, LIU Zhe, LIU Jian-wei
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  82-88.  doi:10.6040/j.issn.1671-9352.2.2014.140
    Abstract ( 539 )   PDF (1953KB) ( 467 )   Save
    References | Related Articles | Metrics
    WLAN has been an essential technology for enterprise Network. However, because of the lack of platform integrity verification mechanism and effective inter-domain access control mechanism, it is difficult to efficiently support emerging applications such as mobile cloud storage.A novel TPM-based WLAN access control scheme was proposed which uses TPM and Attribute-based access control to extend the trust chain from platform to the whole enterprise network and perform fine-grained access control, which ensure that the enterprise WLAN is secure and trusted.
    Matrix description and properties of global avalanche characteristics
    YUAN Hong-bo, YANG Xiao-yuan, WEI Yue-chuan, LIU Long-fei, FAN Cun-yang
    JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2014, 49(11):  89-94.  doi:10.6040/j.issn.1671-9352.2.2014.212
    Abstract ( 551 )   PDF (741KB) ( 561 )   Save
    References | Related Articles | Metrics
    The global avalanche characteristics matrix representation method was proposed by starting from the expression of global avalanche characteristics.And the same absolute value indicator of Boolean functions f(x) and f(x+α) were proved.The relationship between global avalanche characteristics (GAC) and Walsh spectrum was studied by matrix representation and the GAC absolute indicator's limits between a Boolean function and an affine functions. At last, the influence on GAC indicator by modifying sequence of a Boolean function was analyzed. In combination with hill-climbing algorithm, a large number of Boolean functions with good absolute value indicator were achieved via M-MF Bent functions.