You are on page 1of 12

International Journal of Advanced Research in Engineering and Technology (IJARET)

Volume 11, Issue 11, November 2020, pp. 101-112, Article ID: IJARET_11_11_010
Available online at http://iaeme.com/Home/issue/IJARET?Volume=11&Issue=11
ISSN Print: 0976-6480 and ISSN Online: 0976-6499
DOI: 10.34218/IJARET.11.11.2020.010

© IAEME Publication Scopus Indexed

SENTIMENT ANALYSIS FOR AFAAN OROMOO


USING COMBINED CONVOLUTIONAL
NEURAL NETWORK AND BIDIRECTIONAL
LONG SHORT-TERM MEMORY
Megersa Oljira
Lecturer, Department of Information Technology, Ambo University,
Hachalu Hundesa Campus, Oromia, Ethiopia

Kabada Sori
Lecturer, Department of Information Technology, Ambo University,
Hachalu Hundesa Campus, Oromia, Ethiopia

Dr. Karthikeyan Kaliyaperumal*


Associate Professor, Department of Information Technology, Ambo University,
Hachalu Hundesa Campus, Oromia, Ethiopia

Bayisa Natea Sima


Head, Department of Information Technology, Ambo University,
Hachalu Hundesa Campus, Oromia, Ethiopia
*Corresponding Author

ABSTRACT
Sentiment analysis has become the most popular research topic due to its various
application in business, politics, entertainment, however analyzing opinion of people
from short text such as Twitter message and single sentence is quite a challenging task
due to their informality, misspell and semantic error. In this study, we propose character
level multiscale sentiment analysis for Afaan Oromoo using combined Convolutional
Neural Network and Bidirectional Long Short-Term Memory (CNN-Bi-LSTM)
approach. Since there is no standardized and sufficient corpus prepared for Afaan
Oromoo Natural Language Processing (NLP) task including sentiment analysis so far,
we have collected data from two domain, Facebook and Twitter for the experiment.
After collecting data, we removed user names, links, none Afaan Oromoo texts, and any
unnecessary characters. The cleaned data were annotated manually by 4 different
annotators into five class namely, 2 ,1, -2, -1, and 0 which represent very positive,
positive, very negative, negative and neutral respectively. This multi-scale sentiment
analysis provides a more refined analysis, which is vital for prioritizing and comparison
of different opinion. Afterward we performed experiments on the prepared corpus from

http://iaeme.com/Home/journal/IJARET 101 editor@iaeme.com


Megersa Oljira, Kabada Sori, Dr. Karthikeyan Kaliyaperumal and Bayisa Natea Sima

Facebook and Twitter by applying Convolutional Neural Network, Bidirectional Long


Short-Term Memory and combined Convolutional Neural Network-Bidirectional Long
Short-Term Memory with character level word embedding. The experimental result
validate that the proposed model outperforms both CNN and Bi-LSTM in both Facebook
and Twitter dataset. Based on the implemented Facebook dataset we achieved a
promising performance accuracy of 93.3%, 91.4%, and 94.1% for CNN, Bi-LSTM and
CNN-Bi-LSTM respectively. Consequently, we executed twitter dataset and achieved
92.6%, 90.3%, 93.8% for CNN, Bi-LSTM and CNN-Bi-LSTM respectively. The result
suggests the possibility of multi-scale sentiment analysis as well as CNN-Bi-LSTM on
Afaan Oromoo. We have also suggested that the accuracy can be improved by building
standardized and sufficient amount of data set, which was one of the most difficult and
demanding tasks of our work
Keywords: Afaan Oromoo, Sentiment Analysis, Multi-scale, Character Level,
Convolutional Neural Network, Bidirectional Long Short-Term Memory
Cite this Article: Megersa Oljira, Kabada Sori, Dr. Karthikeyan Kaliyaperumal and
Bayisa Natea Sima, Sentiment Analysis for Afaan Oromoo using Combined
Convolutional Neural Network and Bidirectional Long Short-Term Memory,
International Journal of Advanced Research in Engineering and Technology, 11(11),
2020, pp. 101-112.
http://iaeme.com/Home/issue/IJARET?Volume=11&Issue=11

1. INTRODUCTION
The growth of the Internet and the wide explosion of social media, such as Facebook, Twitter,
and YouTube, has formed many new chances for people to express their attitudes towards any
individual, services, organizations, political parties, government policy, etc. by using their own
native languages.
The tracking of these opinions in social media has fascinated an increasing level of
concentration in the research community. Since large numbers of users share their views,
opinions, judges, ideas, and opinions it is hard to handle such a huge amount of online content.
It is a critical need for automated methods of opinion analysis, which allows in a short time to
process large amounts of data and to understand the polarity of users' messages. Sentiment
Analysis also called opinion mining is the task of natural language processing which concerns
with determining the writers, speakers or another subject attitude with respect to a certain topic
or event. The rapid development and the popularity of social media networks results a huge
amount of user generated content to be available online. Determining the polarity of this user
generated content has substantial benefit in different areas. In business, it allows companies to
automatically gather the opinions of their customers about their products or services and
identify areas of improvement[1]. In politics, it can help to infer the public orientation and
reaction towards political events, which helps in the decision making[2].
A multi-scale sentiment analysis gives a better insight to sentiment analysis as it shows the
degree to which a given text is positive or negative. The previous study focuses on binary
classification. However, Interpretation of opinions could be challenging for humans as binary
distinction of opinions as only positive or negative may not good enough[3], [4],[5]. The scale
values signify the strength of positivity or negativity of a text as rank. It provides a more refined
analysis and quick indication of tone, which is vital for numerous real-life application such as
prioritizing and comparison of different opinion. Although sentiment analysis can be useful to
any natural language. Sentiment analysis studies have been conducted most widely for English
language. However, the methods used for English language cannot be directly applied on other

http://iaeme.com/Home/journal/IJARET 102 editor@iaeme.com


Sentiment Analysis for Afaan Oromoo using Combined Convolutional Neural Network and
Bidirectional Long Short-Term Memory

languages due to dissimilarity of its structure. Afaan oromoo Language which is morphological
rich language has no exception.
Afaan Oromoo is one of the Cushitic branches of the Afro-Asiatic language family. In
Ethiopia, more than 35 million Oromo people use as mother tongue and neighboring peoples
by an additional half-million parts of northern and eastern Kenya[6]. Afaan Oromoo is also
widely spoken by other countries like Tanzania and Somalia, which is one of the most resource-
scarce languages[7]. Nevertheless, there is a few computational linguistic related works and its
one of the least researched and under resourced language. As well as Parsing, part of speech
tagging, and named entity recognition are challenging sentiment analysis in the field of natural
language processing (NLP) [8] including Afaan Oromo so far. This issue can be solved using
machine learning and state-of-the-art deep learning techniques because deep learning models
are effective due to their generalizability and automatic learning capability.
In this study we proposed a combination of convolutional Neural Network (CNN) and
Bidirectional Long Short-Term Memory (Bi-LSTM) with character level embedding. In recent
years, many text classification methods based on CNN or RNN have been proposed[9],[10]
,[11] [12]. CNNs are able to learn sequential correlations. In contrast to CNN, RNN are
specialized for sequential modeling but unable to extract features in parallel. So, we proposed
to CNN-BiLSTM to get the advantage of the two algorithms. We performed experiments by
collecting short text from Facebook and Twitter. Moreover, we compared the performance of
CNN, Bi-LSTM separately with the combined CNN-LSTM. As well as we compared the
performance of CNN, Bi-LSTM and CNN-Bi-LSTM on small and huge amount of dataset.

2. LITERATURE REVIEW
In this section mainly discuss sentiment analysis attempts on Afaan Oromoo and deep learning
for sentiment analysis.

2.1. Sentiment Analysis in Afaan Oromoo


Only a few attempts were conducted on sentiment analysis of Afaan Oromoo language. We
have seen only three studies on sentiment analysis of Afaan Oromoo. The author[13], conducted
the first study on aspect-based summarization of Afaan Oromo news text. They used rule-based
approach. According to the report of the author, the method shows good results. Nevertheless,
lack of the linguistic resources such as Part of Speech Tagging (POS) and lexical database made
the work challenging.
Additional study also conducted by [14] using unsupervised method for Afaan Oromo
sentiments mining. They conduct experiment on unigram, bigram and trigram feature. Based
on the experiment of unigram and bigram feature gets high recall and precision respectively.
Another study conducted by[15], on sentiment analysis for Afaan Oromoo using machine
learning approach. Their study was undertaken on the document level sentiment analysis. They
have conducted experiment on Naïve Bayesian, Long short-term memory and convolutional
neural network algorithm. According to the report of researcher shows they get 90.7%, 71.1%,
54.6% for unigram, bigram and trigram result for Naïve Bayesian, and they get 92% and 93.6%
for CNN and LSTM respectively. They reported that the result is promising. However, to the
best of our knowledge there is still no study conducted on character level multi scale sentiment
analysis using combination of CNN-Bi-LSTM for Afaan Oromoo. The lack of resources such
as lemmatization and morphological analysis tool for the language make difficult errors
specially in sentiment analysis.

http://iaeme.com/Home/journal/IJARET 103 editor@iaeme.com


Megersa Oljira, Kabada Sori, Dr. Karthikeyan Kaliyaperumal and Bayisa Natea Sima

2.2. Deep Learning for Sentiment Analysis


The author[9] investigated the first work on Convolutional Neural Network to provide simple
and effective architecture of text classification. After this investigation, various improved
models based on Convolutional Neural Network have been proposed.
Santos and Gatti [16] proposed Sentiment Analysis of Short Texts using deep convolutional
neural networks (CNN) approach. This proposed deep CNN draw information from character
up to sentence level. They proposed Character to Sentence Convolutional neural network
(CharSCNN). This CharSCNN uses two convolutional layers to obtain related feature from any
size of words and sentences. They performed experiment on two corpora, the Standford
Sentiment Trebank (SSTb), which holds movie reviews sentence; and also, the Stanford Twitter
Sentiment corpus (STS), which contain sentence taken from Twitter messages. According to
their report, they achieved 85.7% accuracy for SSTb corpus and 86.4% accuracy for STS
corpus.
In [17] the combination of convolutional neural network and recurrent neural network were
applied for sentiment analysis of short text. They intended to take advantage of the coarse-
grained local features produced by Convolutional Neural Network as input for RNN and long-
term dependencies learned by Recurrent Neural Network for sentiment analysis output. They
conduct experiment on MR, SST1 and SST2 corpora. Finally, they got 82.28%, 51.50% and
89.95% accuracy for MR, SST1 and SST2 corpora respectively.
Additionally, the author [18]proposed Deep CNN-LSTM for IMDb Review Sentiment
Analysis with Combined Kernels from Multiple Branches. The novelty of their proposed
network is having combined kernel through multiple branches that take data and perform
convolution. Finally, their proposed model got 89% accuracy. Review
Moreover, [19]proposed sentiment analysis for movie by combining CNN and Bi-LSTM as
a complex model. They have designed the suitable structure to combine CNN and Bi-LSTM
and to find out the most optimal layer. They have also conducted six experiments for the test
and accuracy comparison including single CNN and Bi-LSTM. However, CNN-Bi-LSTM
achieved 89.7% accuracy which is better than single CNN and Bi-LSTM. Furthermore, CNN
with one convolutional and pooling layer outperformed CNN with more layers.
Likewise,[20] conducted their study on sentiment analysis from twitter by Multi-Channel
lexicon integrated with CNN-Bi-LSTM. Their study was accompanied on the document level
sentiment analysis. They applied multichannel technique on lexicon to improve the feature of
lexicon. They have used word2vec word embedding. They have used SemEval Task 4 (2013-
2016) corpus for experimental purpose and their model achieved 64% F1 score.
The author [21] also conducted their study on document level sentiment analysis using
mixture of Convolutional neural network (CNN) and bidirectional long short-term memory (Bi-
LSTM) with Doc2vec embedding. They wanted to create a model that takes advantage of the
strength of CNN and Bi-LSTM. They performed experiment on dataset collected from national
and international newspapers and achieved 90.66% accuracy.

3. THE PROPOSED SENTIMENT ANALYSIS


3.1. Preparation of Corpus
There is no standard corpus prepared for Afaan Oromoo sentiment analysis. For the study
purpose would be collected 12,000 comments from the Facebook and 12,000 of tweets from
the Twitter. The collected data is annotated by 4 different linguistic experts into five classes
namely, 2 ,1, -2, -1, and 0. This means, very positive, positive, very negative, negative and
neutral respectively.

http://iaeme.com/Home/journal/IJARET 104 editor@iaeme.com


Sentiment Analysis for Afaan Oromoo using Combined Convolutional Neural Network and
Bidirectional Long Short-Term Memory

3.2. Preprocessing
Preprocessing is very import as it reduces the computational time and increases the classifier
performance because noisy data can slow the learning process and decrease the efficiency of
the system. For this reason, to exclude irrelevant data from the dataset and performed different
preprocessing. Accordingly, our preprocessing includes the following:
Cleaning: Removal of user names, Removal of links, Removal of none Afaan Oromoo
texts, unnecessary characters, etc. After preprocessing is completed the training and testing
dataset is chosen.
This followed 80%-20% rule, in which 80 of the datasets is randomly chosen for training
and 20% is for testing purpose.

3.3. Model Analysis


Morphological rich language like Afaan Oromoo challenging for classification. So, considering
at character level, is a robust method for such language as it captures the shape of information
and are capable of detecting the morphological make up of word. In addition to that one of the
big issues of social media texts are misspelled and informal. To tackle this problem, research
proposed the same strategy proposed by[22]. Our datasets contains, the dictionary of the
following 76 characters: abcdefghijklmnopqrstuvwxyz0123456789-
,;.!?:”’/\|_@#$%^&*~`+=<>()[]. Afaan oromoo character is similar with that of English except
apostrophe (‘) and combinational characters such as ch, dh, ny, ph, sh, ts. These combinational
characters are read as a single character. In other case the apostrophe (‘) is part of word in Afaan
Oromoo unlike English language.

Figure 1The Proposed AO Multi-Scale SA

http://iaeme.com/Home/journal/IJARET 105 editor@iaeme.com


Megersa Oljira, Kabada Sori, Dr. Karthikeyan Kaliyaperumal and Bayisa Natea Sima

The sentiment analysis model we proposed is illustrated in Figure 1. The model is developed
by utilizing both Bi-LSTM and CNN[23].
All the sentences are padded with zero to make all the sentences equal size. Then, the model
receives a sequence of characters (a sentence) as input, and then finds the corresponding One-
hot vector for each character through the dictionary which containing l characters. For
characters or spaces that do not appear in the dictionary, we assign a 0 vector to them.
First of all, represent each character in the word of sentence as a vector of dimension .
Let, say  is the length of the character in the word and  is the dimension, it forms the matrix
  . Where, C indicates the matrix, d is the height or dimension of the matrix and l is
simply the number of characters in word w.
The next step is to apply a convolutional filters H. convolutional filter (kernel) convolves
over the matrix and capture information. At a time, the convolutional filters slide over the
window of the matrix with height  is the same with the height of original matrix C and width
w is number of characters that convolutional filter take at a time. This is user defined and could
be smaller than l. The weight with convolutional H is randomly initialized and will be adjusted
during training.     =convolutional filter matrix of width . Now we overlay H on the
leftmost corner of C and take an element wise product of H and its projection on C. Therefore,
all the output matrix is sum up to get a feature map f . This value set as the first element of a
new vector called .          we then slide H one character to the right and
perform the same operations then get the product and sum up all the values in the resulting
matrix to get another scalar. This scalar value is taken as the second element of     
    . we repeat the same operations character by character until we reach the end of
the word. In each step we add one more element of  and lengthen the vector until it reaches its
maximum length which is     . The vector  is a numeric representation of the word
obtained when we look at the word  characters at time, where n is number of characters. One
thing to note is that the value within the convolutional filter H don’t change as H slides through
the word in fancier term we call H “position” invariant. The position invariance of the
convolutional filter enables us to capture the meaning of the certain letter combination no matter
where in the word such combination appears. Maximum pooling function is applied to keep the
maximum value of . This maximum is collected together to form fixed dimensional
representation of parts of word. This process is referred to as “max-pooling”.   
 . Repeating the same operation with different convolutional filters H. This
convolutional filter may have different width. For example, H=2. As the previous filter we slide
along H’ across the word to get the vector  and then perform max pooling on f, get summary
scalar.      we repeat this scanning process several times with different
convolutional filters with each scanning process resulting in one summary scalar.
The output from the previous operation (CNN) is fed to another function called BiLSTM.
The output of this multiple filters after maximum pooling operation is fed into Bi-LSTM
network.

3.4. Bi-LSTM
LSTM was firstly proposed by [24] to overcome the gradient vanishing problem of the
traditional RNN. The key objective is to introduce an adaptive gating mechanism, which
decides the degree to keep the previous state and memorize the extracted features of the current
data input. Given a sequence            where l indicates the length of input text,
LSTM processes the texts word by word. At time-step , the memory cell  and the hidden
state  are calculated and updated with the following equations:
 =               (1)

http://iaeme.com/Home/journal/IJARET 106 editor@iaeme.com


Sentiment Analysis for Afaan Oromoo using Combined Convolutional Neural Network and
Bidirectional Long Short-Term Memory

=          (2)

 =               (3)

 =             (4)


      (5)
where  is the input at the current time-step  and  is the input gate activation, forget
gate activation and output gate activation respectively,  is the current cell state, represents
the logistic sigmoid function and  signifies element-wise multiplication. For the sequence
modeling tasks like sentiment analysis, it is useful to take access to the past context as well as
the future context. [24]proposed BiLSTM to improve the unidirectional LSTM by introducing
a second another hidden layer, where the hidden to hidden connections flow in reverse
sequential order. Therefore, the model is able to capture information from both the past and the
future. In this paper, BiLSTM is utilized to exploit the past and the future information. As
depicted in Figure 1, the network comprises two sub-networks for the forward and backward
sequence context respectively. The output of the  th word is shown in the following equation:

    
  (6)
Here,  indicates element wise summation to combine the forward and backward pass
outputs.
For the input sentence, we set the number of hidden layers as m, the result of BiLSTM
network can be expressed as follows:
      ]
where  indicates the length of the input text. The RNN network result is
   where each row of  represents the feature of one word generated by BiLSTM.

3.5. Output Layer


The final states of the BiLSTM are concatenated and fed into Softmax activation function,
which calculates a probability distribution over a set of classes. Softmax classifier takes the
output at the last step l and  serves as its input. As noted above, given l texts with w words,
we predict the sentiment y for each text, here text indicates comments for Facebook datasets
and tweets for twitter. The actual annotations of texts are represented by  
       . The predicted values  can be calculated by:
        (7)

 =  (8)
We then use the cross entropy to train the loss function. We first derive the loss of each
labeled comment and the final loss is averaged over all the labeled tweets or comments by the
following equation:

Loss =        (9)

where n indicates the n th input comment. Adam optimizer [25] is used to adaptively adjust
learning rate and optimize parameters of the model. At each hidden layer, we also introduce
dropout[26].

http://iaeme.com/Home/journal/IJARET 107 editor@iaeme.com


Megersa Oljira, Kabada Sori, Dr. Karthikeyan Kaliyaperumal and Bayisa Natea Sima

4. EXPERIMENTAL ANALYSIS
In this section, we evaluate the performance of the proposed system on different datasets. The
experiments were carried out using window 10, Python 3.7.3 and TensorFlow 1.13.1

4.1. Dataset Preparation


We conducted our experiment on different datasets collected from different platforms such as,
Facebook and twitter. The collected datasets are annotated separately by different linguistic
experts.

Table 1 Statics of multiscale sentiment analysis of Afaan Oromoo dataset


Neutral
Datasets Positive Negative Very positive Very negative
Total
2400
Facebook 2400 2400 2400 2400
12000
2400
twitter 2400 2400 2400 2400
12000
Total grand 24000

4.2. Parameter Settings


The convolutional filter sizes were set as 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12. We used rectifier
linear unit (ReLU) activation function for CNN-based models, with dropout 0.1 and mini-batch
size 32. In BiLSTM network, the number of hidden layer m was set to 256. The length of each
fully convolutional layer was set to 512 and 1×1 kernel was used. The details of the parameters
are described in the table below.

Table 2 Parameter Settings for CNN


Hyper parameter training parameters
Convolutional filers (kernel size) 1,2, 3,4,5,6,7,8,9,1
dropout 0.1, at fully connected layer
Pooling Max-pooling
Number of filters 64
epochs 80
Learning rate Default (0.001), beta_1=0.9, beta_2=0.999
Batch size 32

Table 3 Parameter Settings for BiLSTM


Hyper parameter training parameters
dropout 0.3
Memory unit 250
epochs 90
Learning rate Default (0.001), beta_1=0.9, beta_2=0.999
Batch size 20

4.3. Evaluation Method


We used testing dataset for evaluation of the effectiveness and efficiency of the proposed
model. In order to fully illustrate the validity of the model, the accuracy, precision, recall and
F-measure evaluation method are used for evaluation of experiments on data sets.

http://iaeme.com/Home/journal/IJARET 108 editor@iaeme.com


Sentiment Analysis for Afaan Oromoo using Combined Convolutional Neural Network and
Bidirectional Long Short-Term Memory

4.4. Result and Discussion


As we have discussed above, CNN-Bi-LSTM has been implemented for this work. We have
trained and tested CNN, Bi-LSTM and CNN-Bi-LSTM algorithms on small and huge dataset
for both Facebook and Twitter. In this section, we have summarized their results and shown
them.

4.5. Experimental Result on Small Dataset


We first used small dataset of 10000 comment from Facebook and 10000 tweets from Twitter,
with a training split of 80% and test 20% for each Facebook and Twitter. We have compared
our method with CNN and Bi-LSTM based models on the selected datasets. The experiment
results are listed in Table 1. As can be seen, the CNN-based models have the better classification
accuracy than Bi-LSTM. Specifically, our method significantly outperforms both CNN and Bi-
LSTM based models. It achieves all best results from both datasets, which validates the
effectiveness of combining CNN and Bi-LSTM into the united model. In addition, the proposed
model inherits the advantages of both CNN-based deep neural networks and Bi-LSTM, which
contributes to the performance improvement of text classification algorithms.

Table 4 Performance of CNN, Bi-LSTM and CNN-Bi-LSTM on small dataset for both Facebook and
Twitter
Facebook Twitter
Models
P (%) R (%) F (%) A (%) P (%) R (%) F (%) A (%)
CNN 89 88.7 88 89.7 89 87 89 88.6
Bi-LSTM 88.5 88 88.6 87.4 88 88.4 86.5 86.3
CNN-Bi-
92.2 92.5 91 90.1 93 94 90 89.8
LSTM

4.6. Experimental Result on Huge Dataset


We further implemented CNN, Bi-LSTM and CNN-Bi-LSTM on huge datasets size to validate
the effectiveness and efficiency of dataset on the performance of proposed model. We used
12000 datasets for each Facebook and Twitter domain, with 80/20 training and testing split rule.
The experiment results are listed in Table 2. From Table 2, we can come to the same conclusion
that the CNN-based models perform better than Bi-LSTM classification models on huge
datasets. Moreover, it can be seen from Table 2 that CNN and Bi-LSTM model get better
performance than small datasets. Furthermore, the proposed CNN-Bi-LSTM model also
performs best among all models on huge datasets, which further validates the effectiveness of
our model as shown in the Figure 2.

Table 5 Performance of CNN, Bi-LSTM and CNN-Bi-LSTM on huge dataset for both Facebook and
Twitter
Facebook Twitter
Models
P (%) R (%) F (%) A (%) P (%) R (%) F (%) A (%)
CNN 94 93.5 93.5 93.3 93 92.5 92.7 92.6
Bi-LSTM 90.4 91 91.6 91.4 91 91.7 90.5 90.3
CNN-Bi-
95 94.8 94.7 94.1 94 94.3 94 93.8
LSTM

http://iaeme.com/Home/journal/IJARET 109 editor@iaeme.com


Megersa Oljira, Kabada Sori, Dr. Karthikeyan Kaliyaperumal and Bayisa Natea Sima

Figure 2 Comparison of the Algorithms

5. CONCLUSION AND FUTURE WORK


In this paper, we proposed multiscale character-level sentiment analysis model for Afaan
Oromoo short text based on combination of convolutional neural network and bidirectional
neural network, which has the ability of efficiently extracting textual semantics. In addition,
due to its simplicity, efficiency of training and better performance in other morphologically rich
languages we used convolutional neural network and bidirectional neural network. Since there
is no standardized corpus prepared for Afaan Oromoo language, we have collected data from
two domain, Facebook and Twitter for the experiment. Nevertheless, based on the compiled
Facebook dataset we achieved.
A promising performance accuracy of 93.3%, 91.4%, and 94.1% for CNN, Bi-LSTM and
CNN-Bi-LSTM respectively. Consequently, we executed twitter dataset and achieved 92.6%,
90.3%, 93.8% for CNN, Bi-LSTM and CNN-Bi-LSTM respectively. Experimental result
validate that the proposed model outperforms both CNN and Bi-LSTM in both Facebook and
Twitter dataset.
The result suggests the possibility of multi-scale sentiment analysis as well as CNN-Bi-
LSTM on Afaan Oromoo. To the best of our knowledge, this study has not conducted
previously and it is essentially one major contribution of our work. We have also recommended
for further researcher, that the accuracy can be improved by building standardized and sufficient
amount of data set, which was one of the most difficult and demanding tasks of our work. Thus,
we can also suggest that it would be interesting to add emoticons, emoji to check the
performance of the model.
In general, our work has significance on Afaan Oromoo language research and usability in
general and on sentiment analysis in particular. The corpus we have developed has a potential
to the future studies in related fields. This study focusses on character level sentiment analysis
using CNN-Bi-LSTM, which used to solve morphological, semantic, misspell and informality
error of word. It also provides advantage in ranking opinion of people in different field.
Furthermore, we have identified different research gaps that we believe could contribute to the
sentiment analysis field on Afaan Oromoo language.

http://iaeme.com/Home/journal/IJARET 110 editor@iaeme.com


Sentiment Analysis for Afaan Oromoo using Combined Convolutional Neural Network and
Bidirectional Long Short-Term Memory

REFERENCES
[1] B. Liu, “Sentiment Analysis and Opinion Mining,” Synth. Lect. Hum. Lang. Technol., vol. 5,
no. 1, pp. 1–167, May 2012, doi: 10.2200/S00416ED1V01Y201204HLT016.
[2] A. Bakliwal, J. Foster, J. Van Der Puil, R. O. Brien, L. Tounsi, and M. Hughes, “Sentiment
Analysis of Political Tweets : Towards an Accurate Classifier,” Proc. ofthe Work. Lang. Soc.
Media (LASM 2013), pp. 49–58, 2013.
[3] B. P. and L. L. Vaithyanathan, “Thumbs up? Sentiment Classification using Machine Learning
Techniques,” in Proceedings of the Conference on Empirical Methods in Natural Language
Processing (EMNLP), Philadelphia, 2002, pp. 79–86.
[4] B. Pang and L. Lee, “Opinion Mining and Sentiment Analysis,” Found. Trends Inf. Retr., vol.
2, pp. 1–135, 2008, doi: 10.1561/1500000001.
[5] B. Pang and L. Lee, “Seeing stars,” in Proceedings of the 43rd Annual Meeting on Association
for Computational Linguistics - ACL ’05, 2005, no. 1, pp. 115–124, doi:
10.3115/1219840.1219855.
[6] H. Kebede, Towards the Genetic Classification of the Afaan Oromoo Dialects. The University
of Oslo., 2009.
[7] W. Tesema and D. Tamirat, “Investigating Afan Oromo Language Structure and Developing
Effective File Editing Tool as Plug-in into Ms Word to Support Text Entry and Input Methods.”
[8] Q. T. Ain et al., “Sentiment Analysis Using Deep Learning Techniques : A Review,” Int. J. Adv.
Comput. Sci. Appl., vol. 8, no. 6, pp. 424–433, 2017.
[9] Y. Kim, “Convolutional Neural Networks for Sentence Classification,” 2011.
[10] S. Liao, J. Wang, R. Yu, K. Sato, and Z. Cheng, “CNN for situations understanding based on
sentiment analysis of twitter data,” Procedia Comput. Sci., vol. 111, no. 2015, pp. 376–381,
2017, doi: 10.1016/j.procs.2017.06.037.
[11] W. Cao, A. Song, and J. Hu, “Stacked Residual Recurrent Neural Network with Word Weight
for Text Classification,” IAENG Int. J. Comput. Sci., vol. 3, pp. 277–284, 2017.
[12] Y. Zhang, M. J. Er, R. Venkatesan, N. Wang, and M. Pratama, “Sentiment Classification Using
Comprehensive Attention Recurrent Models,” pp. 1562–1569, 2016.
[13] W. TARIKU, “Sentiment Mining and Aspect Based Summarization of Opinionated Afaan
Oromoo News Text,” 2017.
[14] J. Abate, M. Meshesha, and A. Ababa, “Unsupervised Opinion Mining Approach for Afaan
Oromoo Sentiments.”
[15] M. O. Rase, “Sentiment Analysis of Afaan Oromoo Facebook Media Using Deep Learning
Approach,” New Media Mass Commun., vol. 90, pp. 7–22, May 2020, doi: 10.7176/nmmc/90-
02.
[16] C. N. dosMa´ıra G. Santos;, “Deep Convolutional Neural Networks for Sentiment Analysis of
Short Texts,” 69 Proc. ofCOLING 2014, 25th Int. Conf. Comput. Linguist. Tech. Pap., pp. 69–
78, 2014, [Online]. Available: https://www.aclweb.org/anthology/C14-1008.
[17] X. Wang, W. Jiang, and Z. Luo, “Combination of Convolutional and Recurrent Neural Network
for Sentiment Analysis of Short Texts,” Proc. ofCOLING 2016, 26th Int. Conf. Comput.
Linguist. Tech. Pap., pp. 2428–2437, 2016.
[18] A. Yenter, “Deep CNN-LSTM with Combined Kernels from Multiple Branches for IMDb
Review Sentiment Analysis,” pp. 540–546, 2017.
[19] Q. Shen, Z. Wang, and Y. Sun, “Sentiment Analysis of Movie Reviews Based on CNN-
BLSTM,” IFIP Int. Fed. Inf. Process. 2017, pp. 164–171, 2017, doi: 10.1007/978-3-319-68121-
4.
[20] J. Yoon and H. Kim, “Multi-Channel Lexicon Integrated CNN-BiLSTM Models for Sentiment
Analysis,” 2017 Conf. Comput. Linguist. Speech Process. ROCLING 2017, pp. 244–253, 2017.

http://iaeme.com/Home/journal/IJARET 111 editor@iaeme.com


Megersa Oljira, Kabada Sori, Dr. Karthikeyan Kaliyaperumal and Bayisa Natea Sima

[21] M. Rhanoui and M. Mikram, “A CNN-BiLSTM Model for Document-Level Sentiment


Analysis,” pp. 832–847, 2019, doi: 10.3390/make1030048.
[22] X. Zhang, J. Zhao, and Y. Lecun, “Character-level Convolutional Networks for Text,” pp. 1–9,
2015.
[23] [23] G. Liu and J. Guo, “Bidirectional LSTM with attention mechanism and convolutional
layer for text classification,” Neurocomputing, 2019, doi: 10.1016/j.neucom.2019.01.078.

http://iaeme.com/Home/journal/IJARET 112 editor@iaeme.com

You might also like