Identifying Topic Modeling Technique in Evaluating Textual Datasets

• 出版年: Lecture Notes on Data Engineering and Communications Technologies
• 第一著者: 2-s2.0-85151958013
• フォーマット: Book chapter
• 言語: English
• 出版事項: Springer Science and Business Media Deutschland GmbH 2023
• オンライン･アクセス: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85151958013&doi=10.1007%2f978-981-99-0741-0_36&partnerID=40&md5=906626bf37657f029d821b55e8579d0e

One of the most popular methods of topic modeling is Latent Dirichlet Allocation (LDA). To date, philanthropic corporate social responsibility (PCSR) activities are ad-hoc in nature, where assistance is provided more to basic needs with very little attention to activities that can contribute to eradicating poverty. Based on previous related literature, it is found that there is no proper categorization and documentation of PCSR-related activities. Therefore, this research is aimed to identify the most suitable LDA approaches for categorizing PCSR activities. The analysis involved five-years data from the annual reports of 19 CSR-award winning companies in Malaysia. For this study, three LDA techniques were considered and compared namely Variational Bayes Inference, Gibbs Sampling and Expectation Maximization. Then, performance measurement was carried out using coherence value and pyLDAvis technique. As a result, the study showed that the LDA Expectation Maximization method is the best topic modelling technique for clustering PCSR documents. Furthermore, this approach can estimate parameters in probabilistic models when dealing with partial, noisy or missing data. The findings offer an insight to be considered by companies in strategizing the CSR activities, particularly philanthropic responsibility in ensuring optimum impact to innovatively support the society. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.