Two of RIT’s Research Scientists, Chikara Hashimoto and Yuki Nakayama, presented their research findings at this year’s Natural Language Processing (NLP) conference. As the annual conference for the Association for National Language Processing (ANLP), it is renowned for the presentation of language process research in Japan and as an avenue for international exchange among scholars and professionals. Rakuten Institute of Technology was a platinum sponsor of the 3-day conference. The conference was held from March 15 to March 19.
Yuki Nakayama’s research paper, アスペクトベース意見分析における日本語評価コーパスの構築（Japanese Benchmark Corpus for Aspect-based Sentiment Analysis）, presents the first standard aspect-based dataset in Japanese in the hotel reviews domain. With greater availability of online reviews, aspect-based sentiment analysis (ABSA), has gained interest over recent years. The study identifies and extracts opinions from text reviews using a dataset collection of 12,000 hotel reviews from Rakuten Travel. The dataset contains over 76,000 review sentences and experiments were conducted using a Bert-based language model approach. Seven aspect categories were annotated, which includes the following: 1.) meal 2.) dinner 3.) breakfast 4.) location 5.) facilities 6.) room and 7.) bath. The dataset study provides an important contribution to understanding customer opinions in the hotel domain, which will benefit further research in the field.
Additionally, Chikara Hashimoto presented his research on methods to automatically extract cases of unexpected benefits from Wikipedia. His presentation derives from research first conducted during his time at Yahoo Tokyo.
RIT would like to thank ANLP for hosting the annual conference, as well as the other sponsors and participants.
Boston Members explore real-world applicability of item classification at NAACL-HTL 2021
Boston Members accepted to Top Tier Conference at NAACL-HTL 2021
Coining Simpson’s Bias and Finding Best Approaches through Data at AAAI 2021
New Data Release on Hotel Reviews using Aspect-Based Sentiment Analysis