چكيده به لاتين
Emotion analysis is one of the sub-branches of natural language processing that
aims to classify texts according to the feelings, beliefs and attitudes expressed in
them. Much research has been done in recent decades to use machine learning
approaches to solve emotion analysis problems, but their main focus has been on
building stronger feature extractors, as the performance of these learners is more
dependent on the choice of how data is represented. In recent years, with the
increase of computational power and the advancement of deep learning science,
with the transfer of pattern learning to the machine, significant advances have
been made in many issues of different disciplines, especially in the classification
of texts and the analysis of emotion. In sentiment analysis, most of these
improvements have been made for English, and in Persian, due to challenges such
as lack of access to sufficient data sets and accurate text processing tools, the
improvements have not been as significant. The purpose of this research is to
design a model based on deep learning and natural language processing that can
understand the process of sentiment analysis with high accuracy. Among the
innovations of the present study, we can mention the precise preprocessing of the
data set, which has been done in the most optimal way possible, which ultimately
has accuracy in deep learning models such as convolutional neural network
and bi-directional long short-term memory neural network. It should be noted that
the convolutional neural network model is one-dimensional and the Relu operator
is used to activate neurons in different layers. The hybrid form is used, which
improves the accuracy of the results compared to the classic deep learning models
such as support vector machine and random forest. It can also be pointed out that
in this study, the neural network used short-term short-term memory, which is
different from the one-way neural network method in that the neurons move back
and forth between layers, which causes they can move both forward and in the
opposite direction between the layers. One of the advantages of this method is the
preservation of important data and information in the underlying layers of the
neural network. The purpose of designing the four models in the present study is
to find the differences in the discussion of classification of comments in each of
the proposed models. From the random forest, the support vector machine, the
convolutional neural network, and the bi-directional long short-term memory
neural network are compared, and finally a comparison is made between the four
designed models, which will show that the deep learning models are significantly
superior to the model. The basics are machine learning, and the ultimate accuracy
of deep learning models shows a advantage in classification. According to
research in the field of sentiment analysis, today the use of basic methods of
natural language processing does not result in high accuracy.
