This module has been designed to provide students with the knowledge and resources required to learn how to extract business value from unstructured data. The topics covered are organized into two main blocks: the first focuses on machine learning and the associated statistical techniques, while the second adopts a more computational perspective, covering natural language processing and image processing, the two most common types of unstructured data.
The module follows a strongly practice-oriented approach while providing sufficient theoretical foundations to ensure that students acquire and consolidate their understanding of both the fundamental techniques and the current state-of-the-art methods.
Professors
The prerequisite knowledge required to take this course includes proficiency in linear algebra, particularly in working with matrices, and advanced statistics, as well as a solid understanding of the fundamentals of artificial intelligence and data analysis techniques. Students are also expected to be familiar with the architecture and management of Big Data pipelines, and to have advanced proficiency in the Python programming language and in the most commonly used notebook development and execution environments, such as Jupyter, Visual Studio Code (VS Code), Anaconda, and Google Colab.nagers.
The objective of this module is for students to acquire the fundamental knowledge and develop sufficient competence in the various techniques for processing unstructured data, with a particular emphasis on natural language. The module also aims to equip students with the necessary criteria to identify the most appropriate methods for analyzing, processing, and extracting maximum business value from this type of data.
- Introduction to Unstructured Data. This session provides a course overview and introduces the challenges associated with handling unstructured data—such as text and images—their prevalence in real-world scenarios, and the importance of analytical techniques for extracting meaningful insights. Foundational concepts from applied neuroscience, standard preprocessing steps, and essential tools for managing unstructured data are also introduced.
- Co-occurrence Analysis and High-Dimensional Data Visualization with PCA. This session examines the frequency and patterns of paired elements (e.g., keywords or codes within a dataset) to uncover associations and structural relationships among data components. It also covers projecting multi-feature datasets into lower-dimensional spaces using Principal Component Analysis (PCA), enabling clearer interpretation of data structure and variance in two or three dimensions.
- PCA (Continued) and Manifold Learning. This session explores a family of non-linear algorithms designed to discover low-dimensional structures embedded within high-dimensional data. By preserving intrinsic geometric relationships, these methods reveal complex patterns that linear techniques like PCA cannot capture.
- Clustering: k-means and Other Models. This session focuses on unsupervised clustering algorithms that partition data into a predefined number of groups. It then extends to probabilistic approaches, modeling data as a combination of multiple distributions to capture more flexible structures and assignment probabilities.
- Clustering (Continued): Interpretation and Selecting the Number of Clusters. This session addresses strategies for determining the optimal number of clusters using likelihood-based criteria that balance model fit against complexity, helping to avoid overfitting and ensure robust, interpretable results.
- Unstructured Data Review. This session revisits the specific challenges of working with unstructured data—particularly natural language and images—their ubiquity in real-world contexts, and the critical role of analytical techniques in deriving actionable insights. Foundational neuroscience concepts, common preprocessing workflows, and key tools for managing unstructured data are also reviewed.
- Rule-Based NLP. This session focuses on rule-based natural language processing methods, which rely on handcrafted patterns and linguistic rules to analyze and manipulate text. Essential techniques such as tokenization, part-of-speech (POS) tagging, named entity recognition (NER), and syntactic parsing are covered. Through practical examples, participants explore the strengths, limitations, and appropriate use cases for rule-based approaches—whether in niche applications or as complements to data-driven models.
- Neural Networks. This session aims to refresh students' understanding of neural networks and prepare them for advanced topics in NLP and deep learning. Key concepts—including perceptrons, activation functions, backpropagation, and core architectures such as feedforward, convolutional (CNN), and recurrent neural networks (RNN)—are reviewed.
- NLP with Deep Learning. This session introduces deep learning approaches to natural language processing, demonstrating how neural networks can tackle complex language tasks such as sentiment analysis, machine translation, and question answering. Techniques including RNNs, Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRUs) are explained, highlighting their advantages over rule-based methods, with practical demonstrations included.
- Embeddings and Vectorization. This session covers the representation of textual data in numerical form for machine learning applications. Word embedding techniques such as Word2Vec and GloVe are introduced, alongside contextual embeddings from models like BERT. Students learn how vectorization methods capture semantic relationships and contextual information, enabling sophisticated language modeling and deeper linguistic understanding.
- Transformers and Generative AI. This session explores the evolution of the transformer architecture. Core concepts—including self-attention and multi-head attention mechanisms—are explained, along with the standard transformer structure and its operational principles. Models such as BERT and GPT are introduced, accompanied by practical examples of their application across diverse domains.
- Generative AI and Business Applications. This session focuses on real-world applications of generative AI in business contexts. Case studies illustrate how AI can enhance customer experiences, streamline workflows, and enable innovative solutions. Ethical considerations, implementation challenges, and best practices for deploying generative AI systems are also discussed. Additionally, emerging architectures such as RAG (Retrieval-Augmented Generation) and Agentic RAG are covered.
- Image Processing: From CNNs to Transformers. This closing session covers the fundamentals of image processing, beginning with Convolutional Neural Networks (CNNs) and their role in tasks such as image classification and object detection. It then addresses the transition to transformer-based architectures in computer vision, demonstrating how these models have surpassed traditional CNNs in tasks requiring contextual understanding and global relationship modeling within images.
The course is delivered through a weekly session divided into two parts. The first part is devoted to a descriptive introduction to the course topics, together with the theoretical and conceptual explanation of those aspects that require mathematical or computational justification. The second part consists of practical sessions, including demonstrations and hands-on activities (both individual and group-based), designed to help students assimilate the material and understand its practical usefulness and application scenarios.
The teaching methodology therefore integrates independent learning, collaborative classroom activities, and continuous formative assessment, ensuring alignment between the learning activities, assessment system, assessment criteria, and the student workload associated with the assigned ECTS credits.
To assess whether students have achieved the intended learning outcomes of the course, a range of assessment activities is used, typically on a weekly basis. Some of these activities are carried out in groups.
| Assessment Type | Weight | Contents | Importance | AIAS Level |
|---|---|---|---|---|
| Attendance and participation | 20% | All course contents | Moderately important | 1 |
| Individual assignments | 40% | Approximately 8 submissions | Very important | 4 |
| Midterm examination | 10% | Content covered up to that point | Moderately important | 1 |
| Final examination | 30% | Entire module content | Very important | 1 |
The assessment criteria apply equally to all students. Students enrolled under the resit scheme are also required to attend classes. Any exceptional circumstances must be communicated to the teaching staff in advance and approved by the academic tutor.
The course is considered successfully completed when the final grade is 5.0 or higher on a 10-point scale.
RESIT POLICY
Students who do not pass the ordinary assessment period will be eligible to take a resit examination during the extraordinary assessment period. To be eligible for the resit examination, all pending assignments and coursework must have been submitted. The final grade for the extraordinary assessment period will be calculated using the same weighting scheme described above, with a maximum possible grade of 6.0/10.
Students who do not attend any of the examinations during the extraordinary assessment period will receive a final course grade of NP (Not Presented).
The following aspects will be assessed:
- The correct application of the concepts taught in class to the proposed exercises.
- The quality of the conclusions drawn from the completed activities and the accurate interpretation of the results obtained.
- The clarity and organization of the presentation of procedures and solutions.
- The appropriate and authorized use of artificial intelligence in the assigned tasks.
- Jurafsky, D., & Martin, J. H. (2022). Speech and Language Processing
- Bird, S., Klein, E., & Loper, E. (2009). Natural Language Processing with Python: Analyzing Text with the Natural Language Toolkit
- Vaswani, A., et al. (2017). Attention Is All You Need.
- L. Tunstall, L.Von Werra & T.Wolf (2022). Natural Language Processing with Transformers: Building Language Applications with HuggingFace.
- Behrouz A., Razaviyayn M., Zhong P., Mirrokni V. (2025). Nested Learning: The Illusion of Deep Learning Architectures.
---