Question Classification using Semantic, Syntactic and Lexical features

Abstract

Question classification is very important for question answering. This paper present our research work on question classification through machine learning approach. In order to train the learning model, we designed a rich set of features that are predictive of question categories. An important component of question answering systems is question classification. The task of question classification is to predict the entity type of the answer of a natural language question. Question classification is typically done using machine learning techniques. Different lexical, syntactical and semantic features can be extracted from a question. In this work we combined lexical, syntactic and semantic features which improve the accuracy of classification. Furthermore, we adopted three different classifiers: Nearest Neighbors (NN), Naïve Bayes (NB), and Support Vector Machines (SVM) using two kinds of features: bag-of-words and bag-of n grams.Furthermore, we discovered that when we take SVM classifier and combine the semantic, syntactic, lexical feature we found that it will improve the accuracy of classification. We tested our proposed approaches on the well-known UIUC dataset and succeeded to achieve a new record on the accuracy of classification on this dataset.

Extracting Compact Sets of Features for Question Classification in Cognitive Systems: A Comparative Study

“…This is the mostly used set of features, since it usually allows obtaining the best results [6][7][8][9][10]. Unigrams are obtained from tokens by applying two operations: i) elimination of tokens labelled with some lexical tags, like DT, IN, and punctuation; ii) sometimes the remaining unigrams are transformed by the stemming into their radical form.…”

“…A benchmark dataset built by Li and Roth [3] is widely used in literature, and NLP techniques able to analyze the question are more or less well established [4]. On the other hand, different authors chose to extract from questions a broad variety of features [5], usually divided into lexical, syntactic and semantic [6,7]. Different works focused on the extraction of particular words, like whword and head-word [8], but good results were gained as long as the number of employed features was increased to a very high number [6][7][8][9][10].…”

A Forward-Selection Algorithm for SVM-Based Question Classification in Cognitive Systems

“…This feature aims at individuating the most informative word of the question for classification purposes. Introduced in [8], it is widely used and recognized to be useful [3,4,6,8,20]. It is extracted here from the parse tree, by using redefined rules, similar to those proposed by Collins [21], already modified in other works [6,8].…”

“…This is the mostly used set of features, since it usually allows obtaining the best results [3,4,6,8]. Unigrams are obtained from the set of tagged tokens of the question, by eliminating tokens with some tags, like DT, IN, and punctuation.…”

“…The UIUC benchmark dataset of classified questions [2] is widely used in literature to train and test QC modules. On the other hand, while NLP methods are quite well established, a broad variety of features, usually divided into lexical, syntactic and semantic [3,4], can be extracted from questions [5]. Support Vector Machines (SVMs) were revealed as the best machine learning method for this context [4][5][6][7], primarily for their feasibility to manage many features, since good results were gained as long as features were employed in a very high number [3,4,[6][7][8].…”