Text-Mining-Driven Review of Recommender Systems and Reinforcement Learning for Building Control and Occupant Interaction
Date:
This invited talk explored the application of text-mining techniques in conducting literature reviews, with a focus on the integration of recommender systems and reinforcement learning for smart building control and occupant interaction. Text-mining was presented as a powerful alternative to conventional literature review methods, enabling the analysis of large volumes of academic publications with improved efficiency and reduced subjectivity.
The talk emphasized the critical shift from traditional manual literature reviews to automated, data-driven methods. Text mining was presented as a transformative tool capable of handling large datasets, identifying research trends, and uncovering insights across interdisciplinary fields.
The methodology outlined in the presentation involved multiple steps, including article retrieval, keyword categorization, pre-processing of text, and the application of Word2Vec for word embeddings. These methods enabled the extraction of semantic relationships and keyword clustering, offering a comprehensive understanding of the research landscape. The results were visualized through heatmaps and clustering diagrams, which highlighted the connections between algorithms, objectives, and data types in the domain of smart building technologies.
The speaker also addressed the limitations of text mining, such as its inability to provide deep contextual analysis and its reliance on the quality of underlying algorithms. Future directions included the integration of transformer-based models, such as BERT, to enhance the comprehension of academic texts and improve the accuracy of insights.
Slide Download: Access the presentation slides here.
If you need further information or specific sections from the detailed methodology in the document, feel free to ask!