Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 43407 (19)
- University: Sharif University of Technology
- Department: Computer Engineering
- Advisor(s): GhassemSani, Gholamreza
- Abstract:
- Temporal relation extraction is one of the challenging research topics in natural language processing semantic level. The purpose of this kind of extraction is to find the temporal ordering between text events so that they can be used in various applications such as question answering and summarization systems.Most of early researches in temporal relation extractionaimed at finding a number of rules and templates for every single temporal relation in English texts. However, with the availability of temporal corpora in English and some other natural languages like Chinese, Korean, Italian, etc., the research trend in this field turned towards the use of machine learning methods. Accordingly, in this dissertation, wedesign and implement the first temporal relation extraction system by using a machine learning method. We first make an annotated temporal corpus for Persian.Then, based on some previous work in English, we implement a base system for temporal relation extraction in Persian texts. The system benefits from a well-known classification tool, i.e., support vector machines (SVMs). We alsoapply different combinations of simple linear kernel to usual features, plus some new features and can earn 60.61 percent accuracy.Then, by adding different tree kernels on dependency trees of the sentences containing temporal relation events to previous system, we reach 64.31 percent of accuracy. Finally, by using the combinations of previous kernels and a new kernel called sequence string kernel, on dependency tree paths between relevant events,we promote the system accuracy up to 66.28 percent
- Keywords:
- Temporal Relation Classification ; Persian Texts ; Automatic Extraction ; Event Extraction ; Information Retrieval
Digital Object List-
محتواي پايان نامه - view
Bookmark