Naslov
Frequently Asked Questions Retrieval for Croatian Based on Semantic Textual Similarity

Autori
Karan, Mladen ; Žmak, Lovro ; Šnajder, Jan

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
Procedings of the 4th Biennial International Workshop on Balto-Slavic Natural Language Processing / - Sofija : Association for Computational Linguistics (ACL), 2013, 24-33

Skup
4th Biennial International Workshop on Balto-Slavic Natural Language Processing

Mjesto i datum
Sofija, Bugarska, 08.08.2013. - 09.08.2013

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
FAQ retrieval ; information retrieval ; semantic textual similarity ; Croatian language

Sažetak
Frequently asked questions (FAQ) are an efficient way of communicating domain-specific information to the users. Unlike general purpose retrieval engines, FAQ retrieval engines have to address the lexical gap between the query and the usually short answer. In this paper we describe the design and evaluation of a FAQ retrieval engine for Croatian. We frame the task as a binary classification problem, and train a model to classify each FAQ as either relevant or not relevant for a given query. We use a variety of semantic textual similarity features, including term overlap and vector space features. We train and evaluate on a FAQ test collection built specifically for this purpose. Our best-performing model reaches 0.47 of mean reciprocal rank, i.e., on average ranks the relevant answer among the top two returned answers.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA