Sentiment and Emotion Analysis of Amazon's Book Review using Long Short-Term Memory and Word-Emotion Association Lexicon

RAYYAN FATHURRAHMAN, Sigit Priyanta, S.Si., M.Kom., Dr.

2022 | Skripsi | S1 ILMU KOMPUTER

Sejak diperkenalkannya Web 2.0, yang telah digunakan dalam pembuatan blog, forum, dan jejaring sosial online, pengguna telah dapat membangun percakapan dan mengekspresikan ide-ide mereka tentang berbagai topik. Analisis sentimen, juga dikenal sebagai penggalian opini, adalah studi tentang perasaan, sentimen, evaluasi, penilaian, sikap, dan emosi orang terhadap berbagai objek, seperti produk, layanan, organisasi, orang, situasi, peristiwa, dan subjek, serta fitur mereka. Selain itu, analisis emosi juga merupakan teknik untuk mengenali jenis emosi manusia yang berbeda, seperti kemarahan, kebahagiaan, dan depresi. Sebagaimana dikemukakan di atas, dalam penelitian ini akan dilakukan analisis sentimen dan emosi. Long Short-Term Memory, subset dari RNNs, dan NRC Word-emotion Association akan digunakan masing-masing untuk melatih model penelitian ini. Pedagang (dalam contoh ini, penulis) dapat memanfaatkan ulasan buku untuk mengidentifikasi kualitas buku dan meningkatkan kualitas toko mereka, sementara pelanggan dapat menganalisis isi buku berdasarkan ulasan, berkat penggunaan analisis sentimen dan analisis emosi. Selain itu, pemilik platform, seperti Amazon, mendapatkan keuntungan dari ulasan karena mereka meningkatkan lalu lintas ke situs mereka. Penelitian ini di LSTM berkinerja baik secara makro atau rata-rata tertimbang. Meskipun dataset berisi sangat sedikit review negatif, model dapat menghasilkan 81,66%, 96,92%, 97,12%, dan 9,63% di rata-rata makro F1-Score, rata-rata tertimbang F1-Score, akurasi, dan kerugian masing-masing. Analisis lebih lanjut dilakukan untuk mendeteksi emosi yang dirasakan pengulas. Ketiga buku di dalam dataset tersebut memiliki emosi dominan yang sama yaitu kepercayaan. Hal ini menunjukkan bahwa ketiga buku tersebut layak untuk direkomendasikan kepada pembaca lain karena resensinya merasa percaya pada buku-buku tersebut. The Hobbit and Divergent memiliki antisipasi sebagai emosi paling dominan kedua dalam distribusi yang dapat disimpulkan bahwa review memiliki beberapa kesamaan dalam genre fiksi. Sedangkan The All The Light We Can't See memiliki kegembiraan sebagai emosi paling dominan kedua. Hal ini memberikan indikasi bahwa para reviewer menyukai cerita perang fiktif sebagai genrenya.

Since the introduction of Web 2.0, which has resulted in the creation of blogs, forums, and online social networks, users have been able to establish a conversation and express their ideas on a variety of topics. Sentiment analysis, also known as opinion mining, is the study of people's feelings, sentiments, evaluations, appraisals, attitudes, and emotions toward various objects, such as products, services, organizations, people, situations, events, and subjects, as well as their features. Furthermore, emotion analysis is also a technique for recognizing distinct types of human emotions, such as fury, happiness, and depression. As stated above, in this research, a sentiment and emotion analysis will be conducted. Long Short-Term Memory, a subset of RNNs, and NRC Word-emotion Association will be used respectively to train the model for this research. Merchants (in this example, writers) may utilize reviews on books to identify the quality of the books and improve the quality of their stores, while customers can analyze the contents of the books based on the reviews, thanks to the use of sentiment analysis and emotion analysis. Additionally, platform owners, such as Amazon, gain from reviews since they enhance traffic to their sites. This research resulted in LSTM performing well in terms of macro- or weighted-average. Although the dataset contains very little on the negative review, the model can result 81.66%, 96.92%, 97.12%, and 9.63% in macro average F1-Score, weighted average F1-Score, accuracy, and loss respectively. Further analysis was performed to detect the emotion that the reviewers felt. The three books inside the dataset have the same dominant emotion which is trust. This indicates that the three books are worth recommending to other readers since the reviewer felt trust in those books. The Hobbit and Divergent has anticipation as the second most dominant emotion in the distribution which can be concluded that the review has some similarity in the fiction genre. While The All The Light We Cannot See has joy as the second most dominant emotion. This gave an indication that the reviewers like fictional war stories as the genre.

Kata Kunci : sentiment analysis, emotion analysis, RNNs, word-emotion lexicon, word embedding.