Classifying unlabeled short texts using a fuzzy declarative approach

A Proposed Multi-Domain Approach for Automatic Classification of Text Documents

“…The last type is using only a readymade ontology, these are ontologies made for a specific domain(s) by different organizations or people, which can be used by researchers saving time and effort as in [1,2,4,8,9,12,19]. Training dataset is being subjected to a readymade ontology in [1,2,4,8,12], with the difference between them in the kind of ontology used.…”

“…In [8] however, two ontologies being used which are WordNet and Wikipedia, the training documents are subjected to both of them forming two lexical chains, one for each ontology. The WordNet is being used differently by [9,19]. While in [19] the query is expanded by being subjected to the WordNet, the category names are entered into the system in [9] then subjected to the WordNet to formulate proximity equations being transformed later on into proximity relations.…”

“…Semi-supervised technique is applied in [5], where both labeled and unlabeled documents exist, and the classifier's decision criterion is learnt based on expectation maximization algorithm. The unsupervised technique used in [3,9,10,13,14,16,19], where ontology is used, to compensated for the absence of the training dataset, with the help of the unlabeled documents, the query or both.…”

“…In transparent systems the interaction between the user and the system is minimum, while in the interaction systems this interaction is crucial as in [10]. In [4,9,19] the hybrid system is used where the system looks like the transparent one but with a little interference from the user in a specific task.…”

“…The first limitation is due to the fact that the precision of the classifier increases with the increased number of training documents. In order to reach higher degrees of precision, additional time, effort and money required for both collecting and labeling training documents, and with the increased number of documents to be labeled, the chance of mislabeling a document increases [9,14]. The second limitation is the difficulty of handling multi-domain documents, where the document is being typically labeled as belonging to only one domain even if it actually belongs to other domain(s) too, that the system cannot recognize.…”