Abstract:
Natural Language Processing, or NLP, is a discipline of computer science that uses computer-based methods to analyze language in text and speech. The purpose and ambition of NLP is to
enable computers to converse quickly and easily. Emailing, texting, and cross-language
communication are all used in our daily lives. Sentiment analysis, also known as opinion mining,
is a branch of natural language processing that mines a text to extract subjective information
from the source material, allowing a company to better understand the social sentiment
surrounding their brand, product, or service by monitoring online conversations. Nowadays, in
the era of social media, firms' service and products have a significant impact on customer input
and feedback. Customers freely express their feelings about a company's service and products on
the company's social media pages. Companies face a significant and tough problem in extracting
meaningful information from this unstructured, unorganized, massive, and fragmented data.
Amhara Media Corporation is one victim of this and sentiment analysis is used for resolving
such issues, it allows businesses to quickly learn what customers think and feel about their
service or product. Sentiment analysis can be performed in Document level, Sentence level and
Aspect level. Document level sentiment classifies the whole document into positive negative or
neutral but the customer opinion need may not in that manner. Hence the negativity or positivity
of a single sentence does not measure the entire document as a whole. Sentence level sentiment
analysis also categorizes the entire sentence as negative, good, or neutral, although consumer
feedback may not be in those categories. Many publications have been written on Amharic
sentiment analysis, however none of them have looked at the aspect level or used a deep learning
approach. This research focuses on sentiment analysis utilizing aspect level with a hybrid deep
learning approach. The Dataset were collected from in Excel format from the Amhara media
corporation's official Facebook page. To improve the quality of the data, the researcher applied
tokenization, stop word removal, Emoji, punctuation, eliminating non-Amharic texts, word
embedding, and normalization and converting English language comments to Amharic and
utilize comment exporter software to collect 10,000. The researcher used CNN, LSTM, Hybrid
CNN with GRU and hybrid CNN with LSTM. Finally, Hybrid CNN with GRU is selected as
best approach due to the better accuracy it registered.
Keywords: Sentiment Analysis, Aspect Level, Amharic, Comment