COMP61332 Text Mining syllabus 2021-2022
This course unit detail provides the framework for delivery in 20/21 and may be subject to change due to any additional Covid-19 impact. Current students should see Blackboard/course unit related emails for any further updates.
We naturally record and communicate much of our knowledge in textual form. However, for many years now the rate of growth of textual information has been such that the individual struggles to keep up to date in his fields of interest. Well before the advent of the Web, people suffered from information overload and information overlook, and today the ease of electronic publication has only exacerbated these problems. It has also long been realised that the vast archives of text at our disposal contain hidden, unsuspected, potentially valuable information: nowhere explicitly stated, but only discoverable through (until recently) serendipity, or painstaking manual identification and linking of often disparate chunks of knowledge.
Text mining has evolved in recent years as a way of mitigating information overload and information overlook, and of helping us discover new knowledge from old. To do this, it employs a battery of techniques from information retrieval, natural language processing and data mining. Although the holy grail of text mining is the discovery of previously unsuspected knowledge, text mining techniques find application in a wide number of areas, to do essentially with the organising, selecting, filtering, combining, association and exploitation of information. Text mining goes far beyond, and is not to be confused with, classic information retrieval (conventional search engine technology).
What makes text mining challenging is the combination of: the core problems of natural language processing (how to make sense of unstructured data (text), how to deal with the ambiguity inherent in language that humans naturally cope with); the problems that arise when dealing with very large amounts of electronic text and the even larger amounts of representations derived from these during processing; the problems of integrating different components, with different input/output specifications and different intermediate representations, in text mining workflows to accomplish sophisticated tasks; and the non-trivial problem of matching actual system capabilities to user expectations and requirements.
Applications of text mining are many and varied: systems to find promising targets for drug discovery, to support systematic reviews, to match CVs to job profiles, to carry out business news analysis for competitive intelligence, to aid discovery of disease-gene associations, to monitor reports of terrorist activity, to help generate hypotheses for scientific research, to direct customer queries to appropriate support staff, to discover positive and negative opinions on topics of interest, to discover hot topics and trends, ...
This course unit aims to provide students with an understanding of principles, issues, techniques and solutions connected with text mining, and to enable them to gain knowledge of how recent advances in text mining relate to innovative approaches to organising, characterising, finding and exploiting large scale textual information in the search for new knowledge.
Introduction: background, motivation, dealing with information overload and information overlook, unstructured vs. (semi-)structured data, evolving information needs and knowledge management issues, enhancing user experience of information provision and seeking, the business case for text mining.
The text mining pipeline: information retrieval, information extraction and data mining.
Fundamentals of natural language processing: linguistic foundations, levels of linguistic analysis.
Approaches to text mining: rule-based vs. machine learning based vs. hybrid; generic vs. domain specific; domain adaptation.
Dealing with real text: text types, document formats and conversion, character encodings, markup, low-level processes (sentence splitting, tokenisation, part of speech tagging, chunking).
Information extraction: term extraction, named entity recognition, relation extraction, fact and event extraction; partial analysis vs. full analysis.
Data mining and visualisation of results from text mining.
Evaluation of text mining systems: evaluation measures, role of evaluation challenges, usability evaluation.
Resources for text mining: annotated corpora, computational lexica, ontologies, computational grammars; design, construction and use issues.
Issues in large scale processing of text: distributed text mining, scalable text mining systems.
A sampler of text mining applications and services; case studies.
15 hours of lectures.
15 hours of labs.
5 hours of consultation
- Oral feedback in class.
- Course Web site.
- Assessment written exam (2 hours)
- Lectures (15 hours)
- Practical classes & workshops (20 hours)
- Analytical skills
- Problem solving
- Analytical skills
- Problem solving
On successful completion of this unit, a student will be able to:
- To compare and contrast methods for sentence segmentation, tokenisation, part-of-speech tagging, syntactic parsing and semantic representation
- To apply techniques such as named entity recognition, entity linking, relation and event extraction to extract information from text, while leveraging resources such as lexical and semantic resources (e.g. Framenet, VerbNet, WordNet), and terminological repositories
- To design and customise text annotation workflows, taking into consideration various annotation formats
- To explain how text mining supports the development of semantic search systems
- To explain the distributional hypothesis, and to compare with each other (1) count-based and (2) compositional distributional semantics models
- To apply various evaluation measures (e.g., Kappa, recall, precision and F-score)
- To investigate methods for social media content analysis
No reading list found for COMP61332.
Course unit materials
Links to course unit teaching materials can be found on the School of Computer Science website for current students.