UPC-USMBA at SemEval-2017 Task 3: Combining multiple approaches for CQA for Arabic

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 275–279, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics UPC-USMBA at SemEval-2017 Task 3: Combining Multiple Approaches for CQA for Arabic Yassine El Adlouni USMBA Fes, Morocco yeladlouni@gmail.com Imane Lahbari USMBA Fes, Morocco imane.lahbari@usmba.ac.ma Horacio Rodr´ıguez UPC Barcelona, Spain horacio@cs.upc.edu Mohammed Meknassi USMBA Fes, Morocco m.meknassi@gmail.com Said Ouatik El Alaoui USMBA Fes, Morocco s ouatik@yahoo.com Noureddine Ennahnahi USMBA Fes, Morocco nahnnourd@yahoo.fr Abstract This paper presents a description of the par- ticipation of the UPC-USMBA team in the SemEval 2017 Task 3, subtask D, Arabic. Our approach for facing the task is based on a performance of a set of atomic classifiers (lexical string-based, vectorial, and rule- based) whose results are later combined. Our primary submission has obtained good results: 2nd (from 3 participants) in MAP, and 1st in in accuracy. 1 Introduction The SemEval Task 3 subtask D, (Nakov et al., 2017), asks, given a query, consisting of a ques- tion, and a set of 30 question-answer pairs, to re- rank the question-answer pairs according to their relevance with respect to the original question. Question Answering, QA, i.e. querying a com- puter using Natural Language, is a traditional ob- jective of Natural Language Processing. CQA dif- fers from conventional QA systems basically on three aspects: The source of the possible answers, that are the threads of queries and answers ac- tivated from the original query, the structure of the threads and the available metadata can be ex- ploited for the task, types of questions include the frequent use of complex questions, as definitional, why, consequences, how to proceed, etc. One fac- tor that makes very attractive the task is that many approaches, rule-based, pattern-based, Statistical, ML, have been applied to face it. See (Nakov et al., 2017) for an overview of frequently used tech- niques. See also the overviews of past contests, (Nakov et al., 2016a) and (Nakov et al., 2016b). 2 Our Approach Due to the negative results in last year participa- tion, for this year we present a system that com- bines different classifiers, going beyond the two classifiers, Arabic and English shallow features- based ones, used last year. The new classifiers fol- low approaches that have produced good results in systems as (Barr´on-Cede˜no et al., 2016), (Mi- haylov and Nakov, 2016), and (Joty et al., 2016). We will refer in what follows to these classifiers as atomic ones and they are further combined for obtaining the final results. 275 The overall architecture of our system is pre- sented in Figure 1. As can be seen, the system performs in four steps, a preliminary step, aim- ing at collecting needed resources, basically Ara- bic and English classified medical terminologies, a learning step, for getting the models, a classifi- cation step, for applying them to the test dataset, and a last step combining the results of the atomic classifiers that are described next. 2.1 Overall description A core component of our approach is the …

Content truncated to 500 words