Overview of Text Mining Expertise @ SCD

(1)

Overview of Text Mining Expertise @ SCD

(2)

Text Mining @ SCD 

Introduction

 Text mining team @ SCD

Started around 2000

Currenty 1 postdoc, 4 PhD students

Tailored, generic text mining analysis

Diverse application areas

Several collaborations and projects.

Supported by more general SCD expertise in a.o.

• Data mining

• Numerical linear algebra

• Optimization

(3)

Strategic mission

 To consolidate, deepen and extend SCD’s text mining expertise

 By combining statistical approaches and domain- specific information

 To support knowledge discovery

through literature analysis in various domains:

Bio-informatics

Knowledge management

Mapping of science and technology

Bibliometrics

(4)

Problem setting

 Given a set of documents,

 compute a representation, called index

 to retrieve, summarize, classify or cluster them

<1 0 0 1 0 1>



<1 1 0 0 0 1>

<0 0 0 1 1 0>

(5)

Problem setting - 2

Information

Retrieval Information

Extraction

Full NLP parsing Shallow

Statistics

Generic Problem

specific

Domain- specific Shallow Parsing

Document analysis &

Extraction of tokens

Text mining goals

Text mining methodology

Overall approach

(6)

Overview

 Bio-informatics

 Knowledge management

 Bibliometrics & scientometrics

(7)

Overview



Bio-informatics

(8)

Document-centered mining

 Given a set of documents,

 compute a representation, called index

 to retrieve, summarize, classify or cluster them

<1 0 0 1 0 1>



<1 1 0 0 0 1>

<0 0 0 1 1 0>

(9)

Gene-centered mining

 Given a set of genes (and their literature),

 compute a representation, called gene index

 to retrieve, summarize, classify or cluster them

<1 0 0 1 0 1>



<1 1 0 0 0 1>

<0 0 0 1 1 0>

(10)

Patient-centered mining

 Given a set of patients (and their records),

 compute a representation, called patient index

 to retrieve, classify them

 ..and/or associate this information to genes

<1 0 0 1 0 1>



<1 1 0 0 0 1>

<0 0 0 1 1 0>

(11)

Functional genomics : gene profiling

 Profile documents, genes, …

using vocabularies (bag of words approach)

 Tailored vocabularies reflect the 'knowledge' of a certain domain:

+ noise reduction (i.e. irrelevant words)

+ direct link with other knowledge bases (eg. Gene Ontology)

vocabulary

T 1 T 3

T 2

gene

Bert Coessens

(12)

Functional Genomics - TXTGate

Distance matrix &

Clustering

Other vocabulary

Bert Coessens; Steven Van Vooren

(13)

Functional genomics – Networks from literature

 gene networks

 term networks

Bert Coessens; Frizo Janssens

(14)

Human genetics

 Collaboration with

Human Genetics Centre @

University Hospital KU Leuven.

 Mining on clinical profile and chromosomal footprint of patients (CGH microarrays)

 Knowledge discovery for genomic annotation

 Aiming at tools and standards for reporting, data entry and visualisation supporting experts in exploring hypotheses in linking phenotypes to genotypes and in inference of novel gene candidates

Steven Van Vooren

Data Analysis Text Analysis

NLP; Ontologies

(15)

Human genetics

 Knowledge discovery for genomic annotation

• From µA-CGH profiles

• From Biomedical text

 Similarity measures for biomedical text

what: patient records, literature, genes, loci, clones why: retrieval, clustering, inference

• Clustering similar patients, genes, loci, documents

• Finding genes associated by patient records

 Extracting entities from text

• gene name symbols, loci, diseases, phenotypes, clinical entities, karyotypes

 Text summarization

• Profiling of patients, genes, loci, clones, clusters of ~ .

Steven Van Vooren

(16)

Overview

 Bio-informatics



Knowledge management

(17)

McKnow Project

Clustering and classification are focal points, as well as scalability because of the huge corpora of available data nowadays.

We incorporate user profiles, and as such regard both users and documents as points in a high-dimensional vector space.

Furthermore, as environments are typically dynamical, care is taken that used methods are easily updatable.

Dries Van Dromme; Frizo Janssens

 Automated and User-oriented Methods and algorithms for knowledge management

 Collaboration with Center for Industrial Management, KUL

(18)

Case studies knowledge management

 Dimensionality of clustered text-mining cases:

 sista papers

• electronically available publications (ps, pdf) – full text

• 1024 x 49.237

 De Standaard

• full text newspaper articles, but a lot of them very short

• 1776 x 39.363 - but much more data available

 kuleuven papers

• electronically available papers pertaining to researchers from different departments (pdf, word,...)

• 576 x 68.257 ! less documents, broader spectrum

 patent abstracts

• international patent abstracts and titles

• 16.488 x 21.019 ! a lot more doc’s, denser spectrum

 PMA papers

• full text publications of the K.U.Leuven dept. of Mechanics

• 380 x 18.206

 Locuslink “known genes with proteins”

• gene documents from MEDLINE abstracts

• 12.263 x 58.924

Dries Van Dromme

(19)

Overview

 Bio-informatics



Bibliometrics & scientometrics

(20)

Scope

 Bibliometrics

the application of mathematical and statistical methods to books and other media of communication

 Scientometrics

the application of those quantitative methods which are dealing with the analysis of science viewed as an information process

 Patent analysis and mining

The analysis of patent information is considered to be one of the best established, directly available and historically reliable methods of quantifying the output of a science and technology system

 Collaboration with Steunpunt O&O Statistieken

<< to consolidate and to further develop Flanders position as a European innovation intensive region >>

(21)

Projects

 1. Domain Analysis

 Mapping of Nanotechnology field from USPTO/EPO patents

• Text-based clustering ; identification of sub-domains

• comparison with IPC (International Patent Classification)

• comparison with FTC (Fraunhofer Technology Classification)

 2. Science-Technology mapping

 link scientific publications (WoS) and new technologies (patents)

• text-based clustering & analysis of citation network structure

• Case study: Ljung

 3. Trend Detection

 assess trends & emerging fields from “change over time” in structure and characterization of clusters & citation network

Dries Van Dromme; Frizo Janssens

(22)

Software

 Preprocessing &Indexing

Lucene & TextPack

 Search engine and webservices

TXTGate and McKnow

(23)

Publications

targetted submissions by Dec

 Bio-informatics (1-2)

(BMC) bioinformatics, special issues,.. (BC)

 More biological journals (BC, SVV)

 Knowledge management (1)

 Scientometrics, SIAM DM,

 Bibliometrics & scientometrics (1)

 Case study  Bioinformatics, Trends in..

 IEEE transactions, engineering, webmining journals

 SIAM DM

High, moderate, fair impact

(24)

Collaborations

 Formalized

 GBOU-McKnow

• partner CIB olv Joost Duflou (Joris Vertommen, Dries Cleymans)

• User Committee (ICMS, Verhaert, LMS, TriSoft, WTCM)

 IWT met Joris V (Steven: aanvullen/corrigeren)

 Steunpunt O&O Statistieken, INCENTIM

• Patent clustering and detection of emerging trends

 Informal

 M-F Moens (SBO ?)

 IBM – Bart VL

 Gasthuisberg en Peter M: TXTGate als ‘vak’

 J&J