Several NLP tasks require the explorations of complex semantic and syntactic phenomena. For instance, in Paraphrase Detection, verifying whether two sentences are valid paraphrases involves the analysis of some rewriting rules in which the syntax plays a fundamental role. In Question Answering, the syntactic information is crucial, as largely demonstrated in (Croce et al., 2011). In these scenarios, a possible solution is represented by the manual definition of an artificial feature set that is able to capture the syntactic and semantic aspects useful to solve a target problem. Determining how to exploit these features to generate robust predictive models is left to the learning algorithm. However, the definition of meaningful features is still a rather expensive and complicated process that requires a domain expert; moreover, every task has specific patterns that must be considered, making the underlying manual feature engineering an extremely complex and not portable process. Instead of trying to design a synthetic feature space, a more natural approach consists in applying kernel methods on structured representations of data objects, e.g., documents. A sentence s<\/em> can be represented as a parse tree that takes into account the syntax of s<\/em>.<\/p>\n<\/div>\n Tree kernels<\/strong>\u00a0(Collins and Duffy, 2001) can be employed to directly operate on such parse trees, evaluating the tree fragments shared by the input trees. Kernel-based learning algorithms, such as Support Vector Machines (Cortes and Vapnik 1995), can operate in order to automatically generate robust prediction models. Different tree representations embody different linguistic theories and may produce more or less effective syntactic\/semantic feature spaces for a given task. Among the different variants discussed in (Croce et al. 2011), we first investigate in this example the\u00a0Grammatical Relation Centered Tree (GRCT) illustrated below<\/p>\n In a nutshell, given a sentence such as “What is the width of a football field ?\u00a0<\/em>” we can use a natural language\u00a0parser, such as Stanford Parser<\/a> to extract a Dependency Parse tree, such as<\/p>\n The parse\u00a0automatically extract linguistic\u00a0information, such as the Part-of-speech tag (e.g., field\u00a0<\/em>is a Noun, whosePart-of-speech tag is NN) or Dependency Relations among words (e.g., football<\/em> in a Noun Modifier of field<\/em>). A Dependency Parse tree can be (quite easily) manipulated and converted in tree structures that can be used by tree kernels defined in KeLP (Croce et al, 2011):\u00a0a GRCT is a tree where\u00a0PoS-Tags are children of grammatical function nodes and fathers of their associated lexicals.<\/p>\n<\/div>\n In order to apply a tree kernel, we first need to load a dataset, where each example contains a\u00a0Tree Representation. The GRCT representation of the previous parsed sentence is:<\/p>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n The GRCT\u00a0representation is thus a representation of a tree with round brackets. Note that, the boundaries for a Tree Representation is |BT| |ET|. To automatically generate GRCT representations (or other tree representations) from text snippets please refer to this page<\/a>. In this example, each question is associated to a class reflecting the aim of the question: in this case the question\u00a0expects an answer with a number, i.e., NUM. The following code, reads a dataset compliant with these representations. Notice that this example is derived from the kelp-full<\/strong> project, from the\u00a0class<\/p>\n and the training \/ test dataset are contained in the folder<\/p>\n So, first we load the datasets with<\/p>\n Then we can define a Partial Tree Kernel (PTK,\u00a0[Moschitti<\/span>, 2006]) with some default parameters (refers to the documentation for details).<\/p>\n In this example, the learning algorithm used is a SVM implementation based on the C-SVM implementation of LibSVM.<\/p>\n The Question Classification task addressed here is a multi-classification problem, where each question has to be associated to one class from a closed set.\u00a0Here, we introduce also the concept of multi-classification. A One-vs-All classifier is instantiated. It must know what is the base algorithm to be applied, and what are the labels to be learned.![\"\"](\"http:\/\/www.kelp-ml.org\/wp-content\/uploads\/2017\/03\/grct.png\")
<\/p>\n![\"\"](\"http:\/\/www.kelp-ml.org\/wp-content\/uploads\/2017\/02\/dep1.png\")
<\/p>\nNUM |BT:grct| (SYNT##root(POS##WP(LEX##what::w))(SYNT##cop(POS##VBZ(LEX##be::v)))(SYNT##nsubj(SYNT##det(POS##DT(LEX##the::d)))(POS##NN(LEX##width::n))(SYNT##prep_of(SYNT##det(POS##DT(LEX##a::d)))(SYNT##nn(POS##NN(LEX##football::n)))(POS##NN(LEX##field::n))))) |ET|<\/pre>\n
\nThe :grct<\/em>\u00a0are used in KeLP to assign a label to each representation. During the kernel definition phase, these labels are used to indicate which representation must be used during the kernel computation.<\/p>\nit.uniroma2.sag.kelp.examples.demo.qc.QuestionClassification<\/pre>\n
kelp-full\/src\/main\/resources\/qc<\/pre>\n
SimpleDataset trainingSet = new SimpleDataset();\r\ntrainingSet.populate(\"src\/main\/resources\/qc\/train_5500.coarse.klp.gz\");\r\nSimpleDataset testSet = new SimpleDataset();\r\ntestSet.populate(\"src\/main\/resources\/qc\/TREC_10.coarse.klp.gz\");<\/pre>\n
\/\/ Define a new Partial Tree Kernel \r\nKernel kernel = new PartialTreeKernel(0.4f, 0.4f, 1, \"grct\");\r\n<\/pre>\n
\/\/ Instantiate a prototype for a binary svm solver\r\nBinaryCSvmClassification svmSolver = new BinaryCSvmClassification();\r\nsvmSolver.setKernel(kernel);\r\nsvmSolver.setCp(3);\r\nsvmSolver.setCn(3);<\/pre>\n
\nThe OneVsAllLearning<\/em> class will be responsible to “copy” the base algorithms, and to perform the learning according to the One-Vs-All strategy.<\/p>\n\/\/ Instantiate a OneVsAll classifier \r\nOneVsAllLearning ovaLearner = new OneVsAllLearning();\r\novaLearner.setBaseAlgorithm(svmSolver);\r\n\/\/ get the labels on the training set List