Incremental hierarchical text clustering with privileged information

Abstract

In many text clustering tasks, there is some valuable knowledge about the problem domain, in addition to the original textual data involved in the clustering process. Traditional text clustering methods are unable to incorporate such additional (privileged) information into data clustering. Recently, a new paradigm called LUPI -Learning Using Privileged Information -was proposed by Vapnik to incorporate privileged information in classification tasks. In this paper, we extend the LUPI paradigm to deal with text clustering tasks. In particular, we show that the LUPI paradigm is potentially promising for incremental hierarchical text clustering, being very useful for organizing large textual databases. In our method, the privileged information about the text documents is applied to refine an initial clustering model by means of consensus clustering. The initial model is used for incremental clustering of the remaining text documents. We carried out an experimental evaluation on two benchmark text collections and the results showed that our method significantly improves the clustering accuracy when compared to a traditional hierarchical clustering method.

Comparative analysis of methods and tools applicability for product and IPSS development based on Text Mining Techniques

“…Clustering methods will be applied in order to distinguish documents which reports practical application of a PSS method or tool from those which have a theoretical perspective. Some approaches for this task are: i) use the BoED as privileged information, being a complement to the textual representation based on the traditional BoW (MARCACINI; REZENDE, 2013;SINOARA et al, 2014); ii) explore the use of other text representation techniques, such as the bag-of-related-words (ROSSI; REZENDE, 2011); iii) employ the interactive textual feature selection (CORRÊA et al, 2015); iv) apply methods to identify descriptors for the clusters found, such as the method proposed by Santos, Rezende and Oliveira (2014).…”

Exploiting feature extraction techniques on users’ reviews for movies recommendation

“…We use the LIHC (LUPI-based incremental hierarchical clustering) [25] for the automatic generation of a hierarchical clustering of texts, and through this, produce a topic hierarchy. This topic hierarchy is then processed so that the most representative topics are used as aspects of the items.…”

“…The technical information used is a traditional bag-of-words representation, containing the frequency of the terms present in the document. Privileged information in text processing domain consists of information besides traditional term frequency (TF) or term frequency-inverse document frequency (TF-IDF) [25]. In this work, we use the part-of-speech tag of words as privileged information, since they represent a linguistic information about the terms located in the documents.…”

Named entities as privileged information for hierarchical text clustering

“…Several state-of-the-art approaches to improve the automatic generation of clusters have been proposed in the literature [20,2,16,6,14]. The LUPI (Learning Using Privileged Information) paradigm, proposed by Vapnik and Vashist [21] to incorporate privileged information in the classification task, was applied to the clustering task in [6] and [14].…”

“…Recently, Vapnik and Vashist [21] proposed a new machine learning paradigm called Learning Using Privileged Information (LUPI), which allows the incorporation of this additional privileged information during a machine learning process. The LUPI paradigm was extended for clustering tasks [14], in which the privileged information is used to obtain a more robust initial clustering model. This initial model is used for incremental clustering of new textual information available, so as to enable its use in dynamic scenarios.…”