Aspect Classification of Social Documents

点击翻译按钮获取中文摘要

The current research project aims at creating tools for analyzing, modelling and classifying social documents into general categories. Two languages are concerned in this research: French and English. Twitter is a popular short messaging service available through Web page, desktop and mobile software. In the current research, Twitter posts will be considered as a corpus of social documents. The main steps of the project are explained through the monolingual social documents classification and the Cross-Language social documents classification. In this research project, various well-known classification models in the machine learning field will be tested and compared to each others such as Naïve Bayes, Random Forest, Support Vector Machine, etc. Moreover, Tweet Natural Language Processing (NLP) tools such as part-of-speech taggers and a dependency parser will be used in order to extract several features based on NLP knowledge of the tweets for the classifiers. Our objective is to identify the optimal combination of features that yields good prediction results, while avoiding overfitting. The Cross-lingual text classification is a major challenge in NLP, since often training data is available in only one language (target language), but not available for the language of the document we want to classify (source language). Classifying French tweets will be more complex and challenging as we do not have affordable Tweet NLP tools for this language in order to apply the same classification method. One can proceed in two ways: First, a monolingual classification method as explained earlier, will be applied on the French social documents, with considering fewer NLP features. The second solution is a Cross-lingual Text Classification Using topic-dependent word probabilities. Having social documents in the two languages, one can adopt a naïve approach by considering the combination of the multiple independent monolingual (and cross-language) text classifiers.