ConvKN at SemEval-2016 Task 3: Answer and Question Selection for Question Answering on Arabic and English Fora

Proceedings of SemEval-2016, pages 896–903, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics ConvKN at SemEval-2016 Task 3: Answer and Question Selection for Question Answering on Arabic and English Fora Alberto Barr´on-Cede˜no†, Daniele Bonadiman‡, Giovanni Da San Martino†, Shafiq Joty†, Alessandro Moschitti†, Fahad A. Al Obaidli†, Salvatore Romeo†, Kateryna Tymoshenko‡, Antonio Uva‡ †ALT group, Qatar Computing Research Institute, Hamad Bin Khalifa University, Qatar ‡Department of Computer Science and Information Engineering, University of Trento, Italy {albarron,gmartino,sjoty,faalobaidli,amoschitti,sromeo}@qf.org.qa {d.bonadiman,kateryna.tymoshenko,antonio.uva}@unitn.it Abstract We describe our system, ConvKN, participat- ing to the SemEval-2016 Task 3 “Commu- nity Question Answering”. The task targeted the reranking of questions and comments in real-life web fora both in English and Arabic. ConvKN combines convolutional tree kernels with convolutional neural networks and ad- ditional manually designed features including text similarity and thread specific features. For the first time, we applied tree kernels to syn- tactic trees of Arabic sentences for a reranking task. Our approaches obtained the second best results in three out of four tasks. The only task we performed averagely is the one where we did not use tree kernels in our classifier. 1 Introduction SemEval-2016 Task 3 challenged the participants on the different steps of the full task of Community Question Answering (cQA).1 Given a set of exist- ing forum questions Q, where each existing question q ∈ Q is associated with a set of answers Cq, and a new user question q′, the ultimate task is to deter- mine whether a comment c ∈ Cq represents a perti- nent answer to q′ or not. This task can be subdivided into three tasks, namely: (A) to assign a relevance (goodness) score to each answer c ∈ Cq with re- spect to the existing question q; (B) to re-rank the set of questions Q according to their relevance against the new question q′; and finally (C) to predict the appropriateness of the answers c ∈ Cq against q′. 1http://alt.qcri.org/semeval2016/task3 Task 3 included these three tasks for English, whereas an adaptation of Task C was proposed for Arabic (Task D). The reader can refer to (Nakov et al., 2016) for a more detailed description of the tasks. Task A was also proposed in the SemEval- 2015 edition (Nakov et al., 2015).2 We designed systems for all tasks. We used the feature vectors designed by Barr´on-Cede˜no et al. (2015) and Nicosia et al. (2015) for tasks A, B and C, whereas we just used a basic feature vector de- rived from the system of Belinkov et al. (2015) for Task D. Most importantly, for tasks A, B and D, we com- bined feature vectors with tree kernels (Moschitti, 2006) for relational learning from short text (Mos- chitti et al., 2007; Moschitti, 2008). In particular, we used the improved models that have been success- ful applied for several tasks and datasets in standard QA, see for example, (Severyn and Moschitti, 2012; Severyn and Moschitti, 2013; Severyn et al., 2013b; Severyn et al., 2013a; Tymoshenko et al., 2014; Ty- moshenko …

Content truncated to 500 words