상세정보

Traditional Open Directory Project (ODP)-based text classification methods effectively capture topics of texts by utilizing the hierarchical structure of explicitly human-built knowledge base. However, they lack entities, important in text classification tasks, and only consider term count based approaches, ignoring the important semantic similarity between words. In this paper, we propose a system to incorporate the DBpedia entities into the Open Directory Project for an improved text classification performance. First, we search for DBpedia entities in the ODP documents. Second, we train a word-entity embedding model, which projects DBpedia entities and words from the ODP into single distributional space. Third, we incorporate the obtained word and entity embeddings into the representation of the ODP categories and documents for the ODP-based text classification. 

1. Introduction 1
2. Background 5
2.1 Open Directory Project 5
2.2 DBpedia 5
2.3 ODP-based Classifier 6
3. Enriching ODP with DBpedia knowledge 8
3.1 Searching for DBpedia entities in ODP documents 8
3.2 Adding DBpedia entities to the ODP documents 9
4. Incorporating Word Embeddings 10
4.1 Word Embeddings 10
4.2 Dual Word Embeddings 11
4.3 Text Classification 12
5. Evaluation 15
5.1 Datasets 15
5.1.1 Training Datasets 15
5.1.2 Test Datasets 16
5.2 Evaluation Metrics 17
5.3 Experimental Setup 17
5.4 Results 19
5.4.1 Parameter Setting 19
5.4.2 Performance Evaluation 19
5.4.3 Qualitative Analysis 22
6. Related Work 25
7. Conclusion and Future Work 27
7.1. Summary of This Thesis 27
7.2 Future Work 27