580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. What are the challenges and limitations of information extraction methods across different types of unstructured big data?
This line of research investigates the effectiveness and limitations of information extraction (IE) techniques when applied to various unstructured data types such as text, images, audio, and video, especially in the context of large-scale (big) data. Its importance stems from the need to transform heterogeneous, high-volume unstructured data into structured formats usable for analytics and decision-making. Understanding these challenges is crucial to improve IE systems' scalability, accuracy, and applicability across multidimensional unstructured datasets.
Key finding: Systematic literature review revealed task-specific and data-type-related challenges in IE subtasks such as named entity recognition and relation extraction for text, visual relationship detection for images, audio event... Read more
Limitations of information extraction methods and techniques for heterogeneous unstructured big data
Key finding: This structured literature review highlights that existing IE techniques are predominantly designed for single unstructured data types and struggle to efficiently process multifaceted unstructured big data. It identifies the... Read more
Key finding: By conducting an exhaustive survey on IE across text, images, audio, and video, the paper synthesizes knowledge on subtasks and associated techniques, illustrating gaps in handling unstructured big data's multidimensionality... Read more
2. How can the usability of unstructured text in big data analytics be enhanced to improve insight extraction?
This research area focuses on understanding and improving the practical usability of unstructured textual data within big data analytics workflows. Researchers recognize that unstructured text presents unique technical and conceptual challenges that degrade usability in analytics contexts. Addressing these issues with models and validation techniques aims to optimize the process from raw data to insightful knowledge extraction by accounting for subjective intentions and contextual needs.
Key finding: Through a systematic literature review and Delphi validation method with industry and academic experts, this work developed and validated a usability enhancement model for unstructured text data. The model identifies key... Read more
Key finding: The paper's identification of challenges in IE subtasks for unstructured text supports the need for usability enhancement by highlighting data preprocessing and transformation limitations. These insights contribute to... Read more
Key finding: The review elucidates the critical role of data preparation, including profiling, matching, format transformation, and cleaning, as foundational processes enhancing unstructured textual data usability for analysis. It... Read more
3. What algorithmic and machine learning methods can reconstruct or extract structure from partially structured or uncertain unstructured data?
This theme addresses the problem of restoring structure—a core challenge when dealing with messy, semi-structured, or corrupted unstructured datasets that lack consistent schemas. Techniques that infer latent table or relational structures using supervised or unsupervised machine learning provide a pathway to convert unstructured data into analyzable formats. This has broad implications where data export or storage processes disrupt original organization, necessitating intelligent recovery mechanisms.
Key finding: The paper proposes the STCExtract algorithm, a two-phase machine learning-based approach for reconstructing table structures and columns from messy, unstructured delimited files. Using clustering algorithms (k-means,... Read more
Key finding: This interdisciplinary study developed computational scoring methods and 33 software functionalities for appraisal of diverse record formats combining archival metrics with data mining techniques. The evaluation showed that... Read more
Key finding: The ORA-SS model advances the semantic representation of semi-structured data by explicitly distinguishing objects, relationships, and attributes including attributes of relationships. This richer schema supports more... Read more