Abstract:
The method of extracting feasible information from expression or text that helps in different
fields to make decision is called sentiment analysis. Sentiment analysis is an important
methodology for all the business organization to improve their service according to
customer’s demand. In this area a lots of techniques has been used for different researches.
We have used deep learning approach to execute sentiment analysis with Bangla dataset in
this work. As a developing country Bangladesh is getting more dependency on online
shopping day by day. People try to judge the overall quality of products or service that are
available on online from reviews of previous customers. In this work our main motive is
to realize perception and organize the customer’s opinion in structured scheme. The main
challenge we have faced at the time of collecting data. Then after doing some necessary
steps we have prepared our dataset appropriate for our model. In this piece of work, we
have used LSTM and combined CNN-LSTM classifiers to find the polarity of a sentence.
We measured the classifiers results in terms of Precision, F-measure, Recall and Accuracy.
In our result it is shown that the current method can calculate better sentiment than previous
method. We can also observe that our applied LSTM gives more accuracy than combined
CNN-LSTM architecture and it achieved 82.54% accuracy. By comparing some results we
can ensure that we have used a significant technique to calculate sentiment of sentence.
The available online platforms of Bangladesh can use our developed model to separate the
reviews according to the polarity of a sentence.
