Explicit Semantic Sentiment Analysis

Organisatorisches

• Art der Arbeit: Masterarbeit/Diplomarbeit
• Interne Betreuer: Dr. Thomas Gottron

Beschreibung

ESA [1] is traditionally used to compute a similarity between text documents by incorporating a reference corpus in order to abstract from the term level. Recent findings [2] showed that ESA actually mines correlation information from the reference corpus. Sentiment Analysis (SA) instead is the task of identifying sentiments in a text, such as positive or negative opinions. Several SA approaches incorporate correlation information to generalize from a few known positive or negative terms. In conclusion ESA would be suitable to incorporate correlation information for feature computation on weighted sentiment dictionaries. This means, a (rather small) vocabulary suitable  for feature generation of sentiments is compared via ESA to the analysed document to assign sentiment labels.

The task would be to analyse and evaluate the applicability of ESA for SA – leading to an Explicit Semantic Sentiment Analysis (ESSA). This entails an analysis of the performance of ESA in dependence of the scenario, the applied reference corpus and the size of the sentiment dictionary.

In more detail, the work should cover:

• Adaptation of ESA to ESSA
• Evaluation of different SA scenarios
• Evaluation of different reference corpora

Requirements

• Good programming skills
• Knowledge of Information Retrieval techniques are of advantage
• Management of large data sets will be necessary

[1] E. Gabrilovich, S. Markovitch. Computing semantic relatedness using Wikipedia-based explicit semantic analysis. In Proceedings of the 20th International Joint Conference on Artificial Intelligence (IJCAI 07), pages 1606–1611, 2007.
[2] T. Gottron, M. Anderka, and B. Stein, "Insights into explicit semantic analysis" in CIKM’11: Proceedings of 20th ACM Conference on Information and Knowledge Management, pp. 1961–1964, 2011.