Budapest University of Technology and Economics, Faculty of Electrical Engineering and Informatics

Ajánlott:
Recommended: Data mining techniques

- Emerging problems in the field of economic analysis at multinational companies. Typical task types in media and text mining.

- Media and text analysis methods, search techniques, indexing, ranking procedures. Bag of words model. Information retrieval models: Boolean model and Vector model. Weighting schemes (tf-idf), cosine similarity. Query optimization. Web search, web mining.

- Text preprocessing steps. Tokenization, stemming algorithms, Porter, Lovins stemmers. Shallow and deep parsing. POS tagging. Syntax trees and dependency graphs. Stanford tools.

- Language detection, language dependence, Zipf's law. NLP (Natural Language Processing) tools.

- Named entity recognition, relation extraction from text. Typical approaches to relation extraction: co-occurrence, pattern matching methods, supervised machine learning methods. Opinion mining as a modern tool of market research.

- Use of deep neural networks in text analysis (LSTM - Long Short-Term Memory) and image and video content analysis (CNN - Convolutional Neural Network).

- Media classification for images and videos. Preprocessing steps. Types and methods of media classification. CBIR (Context-Based Image Retrieval), simple image processing procedures.

- Connecting image and text modalities. Deep learning methods and systems. Generative Adversarial Network (GAN).

- Reduction of the problem space of text corpora and media datasets, feature extraction and feature selection techniques.

- Text classification. Types and methods of text classification. Naïve Bayes classifier. Rocchio algorithm. Automatic text processing (text generation with deep learning).

- Chatbots and virtual assistants used in companies.

- Cost-effective classification. Active learning. Ensemble classifiers. Clustering media and text datasets.

- Single-label and multi-label text classification. Change tracking in classification tasks. Concept drift.

- Media recommendation systems.

The programming language used in the labs is Python with the appropriate program libraries.

Laboratories:

- Calculation of a weighting scheme (tf-idf) for a text corpus.

- Text preprocessing, indexing, stemming.

- Sentiment analysis.

- Digit recognition task with Keras program library.

- Application of deep learning methods.

- Effective classification (on text corpora, media datasets).

- Text mining application in the economic field.

a. During the teaching period: a midterm test.

b. During the exam period: preparing the solution to the homework (started in the teaching period) (written), defending it in the exam (oral).

c. The condition for signature is to write at least to a sufficient level in the midterm test (including repeated midterm tests: see in the next point). The midterm test or repeated midterm test is successful if the student has reached the maximum score of at least 40%.

d. At least five of the laboratory exercises must be successfully completed to obtain a signature.

The exam consists of 2 parts: an oral examination of the entire course material of the semester and a defense of the previously submitted written homework. The 2 parts count 50-50% in the final mark.

Blanken, de Vries, Blok, Fres (eds): Multimedia Retrieval. Springer, 2007.

Christopher D. Manning, Prabhakar Raghavan, and Hinrich Schütze: Introduction to Information Retrieval. Cambridge University Press, 2008

Ronen Feldman, James Sanger: The Text Mining Handbook: Advanced Approaches in Analyzing Unstructured Data, Cambridge University Press, 2007