CONTENIDO
1. Introduction 1
1.1. Motivation 1
1.2. Objectives 3
1.3. Outline of thesis proposal 4
1.3.1. Text Analysis 4
1.3.2. Drug name recognition 4
1.3.3. Anaphora resolution 5
1.3.4. Extraction of Drug-Drug Interactions 5
1.4. Document Structure 5
2. Evaluation of Biomedical Information Extraction Systems 7
2.1. Methodologies 7
2.2. Evaluation Measures 8
2.3. Unsolved issued in evaluation process 9
3. DrugDDI: an annotated corpus for Drug-Drug Interaction Extraction 11
3.1. Biomedical corpora for relation extraction 11
3.1.1. Open Issues on biomedical corpora for relation extraction 13
3.2. The DrugDDI corpus 14
3.2.1. Collecting the corpus 14
3.2.2. Processing the corpus 15
3.2.2.1. Failure Analysis of MMTx 20
3.2.3. Annotating the corpus 21
3.2.3.1. DDIAnnotate tool 24
3.3. Conclusions 25
4. Drug Name Recognition and Classification 27
4.1. Introduction 27
4.2. Biomedical Named Entity Recognition 29
4.2.1. Dictionary-Based Approaches 29
4.2.2. Pattern-Based Approaches 30
4.2.3. Machine Learning Approaches 32
4.2.4. Unsolved Issues in Drug Name Recognition 33
4.3. DrugNer: drug name recognition and classification 34
4.4. Evaluation 36
4.5. DrugNer Viewer tool 38
4.6. Conclusions 39
5. Anaphora Resolution for Drug-Drug Interaction Documents 41
5.1. Introduction 41
5.2. Related Work in biomedical anaphora resolution 42
5.3. Building a corpus to support the anaphora reference resolution for Drug-Drug Interactions 44
5.4. Identification of anaphoric expressions 46
5.4.1. Identifying pronominal anaphora 46
5.4.2. Identifying drug nominal anaphora 46
5.5. Scoring-based method for resolving antencedent candidates 48
5.6. Linguistic rules-based method for resolving antecedent candidates 49
5.6.1. Determination of the anaphora scope 49
5.6.2. Antecedent selection 49
5.7. A baseline for drug anaphora resolution 50
5.8. Experiment results of the anaphora resolution 51
5.9. Conclusions 52
6. Related work for Relation Extraction in the biomedical domain 55
6.1. Introduction 55
6.2. Linguistic-based approaches 57
6.3. Pattern-based approaches 58
6.4. Machine learning approaches 61
6.4.1. Features-based methods for Relation Extraction 61
6.4.2. Kernels-based methods for Relation Extraction 63
6.5. Unsolved Issues in Biomedical Relation Extraction 69
7. Combining syntactic information and patterns for Drug-Drug Interaction extraction 73
7.1. Introduction 73
7.2. Detecting coordinate structures 74
7.3. Identifying appositions 77
7.4. Clause splitting 78
7.4.1. Rules for Sentence Simplification 84
7.5. Evaluating syntactic structures resolution 85
7.6. The set of lexical patterns to extract DDIs 85
7.7. Evaluation 88
7.8. Conclusions 91
8. Using a kernel-based approach for Drug-Drug Interaction extraction 93
8.1. Introduction 93
8.2. A shallow syntactic kernel for relation extraction 95
8.2.1. Global Context Kernel 95
8.2.2. Local Context Kernel 96
8.2.3. Shallow Linguistic Kernel 98
8.3. Evaluation 98
8.3.1. Datasets 99
8.3.2. Experimental results 102
8.4. Conclusions 106
9. Conclusions 109
9.1. Conclusions and Future Work 109