Improving multi-choice question answering by identifying essential terms in questions
Journal Title: Revista Romana de Interactiune Om-Calculator - Year 2018, Vol 11, Issue 2
Abstract
This paper describes a method for identifying the most relevant or essential terms in a question: the words that define the meaning of the question and whose removal makes it impossible to be answered correctly even by human agents. We use an artificial neural network architecture built upon semantic and syntactic features extracted from the question. The neural network prediction represents the degree to which the given term is essential to the question. The model has been trained and validated on a dataset consisting of 2233 questions and about human 18000 labeled terms with essentialness information. It achieves an F1 score of 0.80 which is similar to other state-of-the-art approaches but requires fewer features (15 compared to 120 features used by similar works) and is much easier to train. We further show how using only essential term information can improve the accuracy of a multi-choice question answering system, based on standard information retrieval (IR) techniques, by up to 4%.
Authors and Affiliations
George-Sebastian Pirtoaca, Stefan Ruseti, Traian Rebedea
Keywords
Related Articles
- EP ID EP673684
- DOI -
- Views 148
- Downloads 0
Creating enhanced interfaces for cyber-physical-social systems: the remote drone control experiment
In the area of cyber-physical-social systems there is always a huge demand for integration and adaptation; most often, these two tasks go together. There are very few practical cases when every module fits perfectly into...
Usability Evaluation of Gesture-Based Interactions using a Smart Watch
This paper describes the testing methodology applied and the results obtained in evaluating the usability of a system for gesture recognition based on a smartwatch and a mobile phone. The data acquisition is performed us...
Criminal detection – O aplicaţie pentru identificarea infractorilor
Prezenta lucrare îşi propune să vină în ajutorul celor ce doresc să identifice rapid un posibil infractor, pe baza informaţiilor, pe care le avem disponibile la un moment dat. Sistemul construit foloseşte ca intrare info...
Goal-oriented conversational agents – a proposed approach for practical domains
Conversational agents are nowadays desired in many fields, mainly for task automation, but also for entertaining, therapy or other purposes. This report will introduce the main Dialogue Systems categories, along with som...
Recrutare web 2.0
Articolul descrie primii paşi în realizarea unui sistem semantic de management al competenţelor. Sistemul este destinat îmbunătăţirii activităţii unei companii de recrutare specializate pe domeniul IT şi utilizează tehno...