Sobkowiczet al.(2012) have used the NLP and the Sentiwordnet dictionary. The purpose oftheir work is to track the behavior of the users about a particular topic onthe social networks.
The proposed model in this article also identifies thesocial structures, types of the communications and the specific individuals.Hridoyet al.(2015) have used the Sentiwordnet dictionary in a feature-based research.In this paper, the sentiment analysis is carried out in certain places usingthe demographic characteristics such as being male and female. To analyze thedata, the tweets in seven cities have been explored about the features of theiPhone 6 mobile phones. Abbasi et al.(2008) is an instance of thestudies conducted in the field of studying with an observer. This paper focuseson the recognition of the common, syntactic and independent language features.
These features are derived from the Arabic and English associations of geneticalgorithms.Thefirst study done on Persian language was conducted by Mohammad Reza Shams et al.(2012)who used the LDA’s unattended method in order to learn the model. The datasetused in this article is 400 comments on the mobile phone products that are categorizedinto two groups of 200.
Nekah et al.(2014)used the method of learning based on the sentence-based monitoring in theirresearch. In this study, a method is proposed in order to define the semantic labelsand combine them with the syntax tags to have a better understanding of theexpressed feelings. Vamerzani & Khademiconducted a research analyzing the users’ reviews, visualizing the results andenhancing the business intelligence using an analysis of the opinions andconsidering the challenges of Persian language in its preprocessing.
They usedan algorithm with a backup vector machine and the TFIDF feature selectionmethod. The F criterion in this study is 90.15 (Vamerzani & Khademi, 2015).Maas etal.
(2011) consider a vector of the words in the space of the multi-dimensionalrelationships. In this paper, they examine the data on a film and use thecombination of LDA and LSA, two observer and non-observer algorithms, in orderto overcome the syntactic and semantic relations simultaneously.Chen & Qi use the machine learning algorithms toexamine the impact of the social networks on the customer purchases. In thispaper, the linear model of CRFs is used for opinion mining. The results revealthat it functions better than the other algorithms of machine learning. It hasdeveloped, for the first time, a three-dimensional architecture in order toinfluence the users’ decision-making process.
It tracks the users’ decisionmaking and behaviors, shows the users a matrix comparison for bettercomparisons of the products, and has given more weight to the negative opinions(Chen & Qi, 2011).Many ofthe proposed methods for extracting the features in a domain require theeducational data related to that specific domain. Therefore, we use the doublepropagation method in order to extract the features(Qiu, Liu, Bu, & Chen, 2011). Due tothe lack of a comprehensive Persian dictionary, we use the translation of thecomprehensive Bing Liu Dictionary, which contains the general vocabularycontaining feelings in English.In theresearches carried out so far, especially on the Persian language, lessattention has been paid to solving the problems of the pre-processing stages.The quantitative researches have focused on extracting the product features anddetermining the polarity of these features.
In the double propagation methodused in research (Golpar-Rabooki et al., 2015), onlythe adjectives are considered as the opinion words, while some verbs alsoinclude the role of a feature or the opinion words. For example the verb”use” is an feature in the sentence “the ease of use”, and theword “liked” in the sentence “I liked the quality of thisphone” contains opinion. To better understand the features and the opinionwords, the verbs and the adverbs that have these features should be identified.
One of the disadvantages of this method is that for determining the polarity ofthe sentences whose opinion words are not polarized, the polarity of thesentences before and after them is considered in order to determine thepolarity of the current sentence. Better methods can be used to determine thepolarity of these sentences.In orderto increase the accuracy of classifying the opinions, it is possible toeliminate the additional and the neutral words in determining their polarity.
