K-Means Clustering for Answer Categorization on Latent Semantic Analysis Automatic Japanese Short Essay Grading System

Automatic Essay Scoring: A Review on the Feature Analysis Techniques

“…The most straightforward example of this analysis is the string-based similarity between words which can be identified through similarity measures such as Cosine and Jaccard. In addition, sometimes the morphology of words could be extended to consider the ASCII code representation of characters within the essay answer ( [38]; [39]; [40]). The way of adopting this feature analysis into an unsupervised technique is simply represented through the use of clustering where similarity values between answers (produced by Cosine or Jaccard) are being used to aggregate similar answers in a single cluster [18].…”

“…The utilization of these approaches for supervised AES is simply represented by feeding a regression technique with answer document vectors [38]. Otherwise, a distance measure such as Cosine can be used to determine similarity between answers documents through an unsupervised technique ([2]; [6]; [18]; [40]).…”

“…Another type of features depicted in the literature is the ones focused on morphological aspect of the words within the essay. This has been represented by utilizing the string-based similarity ( [18]; [39]; [40]). On the other hand, the statistics of words within the essay have been also utilized ( [12]; [21]).…”

“…Yet, the problem of such external knowledge source is that it can provide general synonyms of the words where the aim sometime is to focus on specific domain. On other hand, some studies attempted to include the semantic analysis without the use of any external knowledge source, but rather through corpus-based approaches ([2]; [6]; [18]; [40]). These approaches are being fed with specific corpus to analyze the similar contexts which reflects on finding semantic correspondences.…”