InfoFramework

Using findings in the data mining, we composed InfoFramework, a framework for experimentation and rapid prototyping utilized for setting up classification experiments.

InfoFramework is designed as modality-, domain- and language-independent and can be used, for example, in discourse (NLP) analysis, in neurobiological processing, forensics, sociolinguistics, in computer games, in analysis of social interaction.

InfoFramework implements means for generating, evaluating, fusing, optimizing statistical datasets as well as creating and testing sample applications that rely on the composed datasets. It defines the core engine and a set of utilities and can be run either as a GUI in a manual mode, or automatically in a batch mode.

The framework was used to implement Emotext for long texts and tested on the following English corpora: Pang corpus, Berardinelli movie review corpus, a corpus with spontaneous dialogues (the SAL corpus), and a corpus with product reviews.

Scientific foundation

Theoretical basis for analysis is laid by a dissertation on opinion mining and lexical affect sensing. It is available in bookshops of 25 countries on five continents.

The dissertation describes an interdisciplinary study that considers linguistic and psychological findings to perform computer-aided categorization of opinions and emotions in texts. It discusses various emotional corpora (movie reviews, weblogs, product reviews, and natural-language dialogues) and describes different approaches to affect classification of their texts: a statistical approach that utilizes lexical, deictic, stylometric, and grammatical information; a semantic approach that relies on emotional dictionaries and on deep grammatical analysis; a hybrid approach that combines the statistical approach and the semantic approach.

Furthermore, this thesis explores affect sensing using multimodal fusion that utilizes lexical and acoustic data. In conclusion, the thesis discusses significant contributions and describes future work.