KLUEless: Polarity Classification and Association

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 619–625, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics KLUEless: Polarity Classification and Association Nataliia Plotnikova and Micha Kohl and Kevin Volkert and Andreas Lerner and Natalie Dykes and Heiko Ermer and Stefan Evert Professur f¨ur Korpuslinguistik Friedrich-Alexander-Universit¨at Erlangen-N¨urnberg Bismarckstr. 6, 91054 Erlangen, Germany {nataliia.plotnikova, micha.kohl, kevin.volkert, andreas.lerner, natalie.dykes, heiko.ermer, stefan.evert}@fau.de Abstract This paper describes the KLUEless system which participated in the SemEval-2015 task on “Sentiment Analysis in Twitter”. This year the updated system based on the developments for the same task in 2014 (Evert et al., 2014) and 2013 (Proisl et al., 2013) participated in all five subtasks. The paper gives an overview of the core features extended by different ad- ditional features and parameters required for individual subtasks. Experiments carried out after the evaluation period on the test dataset 2015 with the gold standard available are in- tegrated into each subtask to explain the sub- mitted feature selection. 1 Introduction The SemEval-2015 shared task on ”Sentiment Anal- ysis in Twitter” (Rosenthal et al., 2015) is a rerun of the shared task from SemEval-2014 (Rosenthal et al., 2014) with three new subtasks. While subtasks A and B were identical to the tasks of SemEval-2014 and dealt with the identification of polarity in a given message, subtask C, D and E were new. In subtask C a topic was given, towards which the sentiment in a message had to be identified. Subtask D was sim- ilar to subtask C, as the sentiment towards a given topic had to be identified, but in this subtask several messages were given from which the sentiment had to be drawn. Ultimately in subtask E, the sentiment of a given word or phrase had to be measured on a score ranging [0, 1], indicating its association with positive sentiment. The training data for subtasks A and B are the same as in SemEval-2014 (Rosenthal et al., 2014) and SemEval-2013 (Nakov et al., 2013). For subtask A, there are 9,505 training items with 6,769 items in development set and 3,912 items in the test set. For subtask B, there are 10,239 training items, 5,907 items in the development set and 3,861 in the test set. For subtasks C and D the same training sets as for subtasks A and B were used by our team. A pilot task E aimed at evaluation of automatic methods of generating sentiment lexicons had no training set, a detailed approach used for this subtask will be given in Section 3. This paper describes the updated system with our efforts to improve it after the evaluation period. The KLUEless system was ranked within the top 3 par- ticipants to subtasks A (rank 2 out of 11), C (rank 2 out of 7) and D (best result out of 6 teams). It scored 5th place in subtask E, but only 13th place in sub- task B (rank 13 out of 40 teams). In the following chapters, we will …

Content truncated to 500 words