Fine-grained opinion analysis is one task in the field of sentiment analysis/opinion mining. This involves, among other things, detecting opinionated subparts of a text and linking targets of opinion and opinion holders together. As a part of the master's thesis, we present an environment for experimentation with the MPQA Opinion Corpus and a variety of linguistic analysis tools and target representations. We will use this to investigate quantitative and qualitative effects when using different types of interface representations to syntactic analysis.
On this project github site, we provide the source code that we developed for the project. In the following, we will give an description of the installation and the usage of these tools.
(The information on this page will be expanded)
After cloning the repository, a config file should be created, overriding the information in the default settings file, masters_project-default.cfg.
We provide in the folder preprocessed_files the preprocessed files for error analysis and experimenting with features.
To use the system with the preprocessed files, set the path masters_project.cfg to the folder with the preprocessed files.
We give three examples of usage:
To run the full system, a number of tools need to install:
- Repp
- Resa
- Stanford POS tagger
- MPQA Opinion Corpus
- Subjectivity Lexicon (Wilson, Wiebe, and Hoffmann)
- LTH SRL parser
- Bohnet & Nivre parser and models
- Liblinear
- Wapiti