The emergence of new and challenging text mining applications is demanding the development of novel text processing and knowledge extraction techniques. One important challenge of text mining is the proper treatment of text meaning, which may be addressed by incorporating different types of information (e.g., syntactic or semantic) into the text representation model. Sentiment classification is one of the challenging text mining applications. It may be considered more complex than the traditional topic classification since, although sentiment words are important, they may not be enough to correctly classify the sentiment expressed in a document. In this work, we propose a novel and straightforward method to improve sentiment classification performance, with the use of semantically enriched information derived from domain expressions. We also propose a superior scheme for generating these expressions. We conducted an experimental evaluation applying different classification algorithms to three datasets composed by reviews of different products and services. The results indicate that the proposed method enables the improvement of classification accuracy when dealing with reviews of a narrow domain.

In this paper, the Filtered-Association Rules Network (Filtered-ARN) is presented to structure, prune, and analyze a set of association rules to construct candidate hypotheses. The Filtered-ARN algorithm selects association rules with the use of asymmetric objective measures, Added Value and Gain, then builds a network allowing more exploration information. The Filtered-ARN was validated using three datasets: Lenses and Soybean Large, both available online for a text and a real dataset with data on organic fertilization (Green Manure). The results were validated by comparing the Filtered-ARN with the conventional ARN and also comparing the results with the decision tree. The approach presented promising results, showing its ability to explain a set of objective items and the aid to build more consolidated hypotheses by guaranteeing statistical dependence with the use of objective measures.

Modern agribusiness management incorporates instruments for risk management with the objective of mitigating uncertainties to the producer. In this context, the producer (risk averse) transfer the risk of price oscillation to companies or individuals that operate in the futures market and who expect to receive a payment (risk premium) for assuming such risk. Defining the adequate strategies for risk management depends on the knowledge about the problem to determine prices ranges in the future. Recent studies demonstrate that time series forecasting can be significantly improved by considering additional information about the problem. In particular, besides the historical time series, textual knowledge extracted from the news portals, social networking and other public data sources available in the web may also be used. This paper presents an approach for agribusiness time series forecasting that allows incorporating external knowledge in the form of events extracted from news about agribusiness, without the need to previously label textual information. In this case, periods of significant uptrends and downtrends of time series are automatically identified - known in the literature as perceptually important points (PIP). We extend the concept of PIP to news events, where similar events published with a certain regularity in periods of uptrends and downtrends are selected as perceptually important events to improve time series forecasting models. An experimental evaluation based on price prediction on ten corn futures contracts (derivatives) provides evidence that the proposed approach is promising.

The discovery of knowledge in textual databases is an approach that basically seeks for implicit relationships between different concepts in different documents written in natural language, in order to identify new useful knowledge. To assist in this process, this approach can count on the help of Text Mining techniques. Despite all the progress made, researchers in this area must still deal with a large number of false relationships generated by most of the available processes. A semantic approach that supports the understanding of the relationships may bridge this gap. Thus, the objective of this work is to support the identification of implicit relationships between concepts present in different texts, considering the verbal semantics of relationships. To this end, analysis based on association rules were used together with metrics from complex networks and a verbal semantics approach. Through a case study, a set of texts from alternative medicine was selected and the different extractions showed that the proposed approach facilitates the identification of implicit causal relationships.

A recommender system can be defined as an information filtering technology which can be used to output a ranking of items (e.g. products, places, etc) that are likely to be of interest to a user. Context-aware recommender systems makes recommendations by incorporating contextual information into the recommendation process. However, there is a lack of automatic methods to obtain contextual information for such systems. In this work, we have proposed to apply clustering techniques to transform geo-referenced data (i.e. latitude and longitude) in contextual information (i.e. regions) to feed the contextual systems. We have evaluated our proposal in the Yelp dataset, which showed evidences that our contextual information can provide better recommendations.

Recommender systems aim to assist users by recommending items that may be of interest to them. Traditionally, these systems use only user and item information. Over time, new information is being used, such as contextual information, which has improved the accuracy of the generated recommendations. In this work, we propose a context-aware recommender method that extracts contextual information from textual reviews using a word embedding based model. In addition, we propose two ways of considering textual contexts in recommender systems, the "Context of Reviews" and the "Context of Items". We evaluated our proposal by using the Yelp dataset (RecSysChallenge 2013); three baselines; and four context-aware recommender systems. In general, our proposal seems to be superior to the three baselines, mainly considering the "Context of Items", and the results were promising, allowing some lines of future work.

Transductive classification is a useful way to classify texts when labeled training examples are insufficient. Several algorithms to perform transductive classification considering text collections represented in a vector space model have been proposed. However, the use of these algorithms is unfeasible in practical applications due to the independence assumption among instances or terms and the drawbacks of these algorithms. Network-based algorithms come up to avoid the drawbacks of the algorithms based on vector space model and to improve transductive classification. Networks are mostly used for label propagation, in which some labeled objects propagate their labels to other objects through the network connections. Bipartite networks are useful to represent text collections as networks and perform label propagation. The generation of this type of network avoids requirements such as collections with hyperlinks or citations, computation of similarities among all texts in the collection, as well as the setup of a number of parameters. In a bipartite heterogeneous network, objects correspond to documents and terms, and the connections are given by the occurrences of terms in documents. The label propagation is performed from documents to terms and then from terms to documents iteratively. Nevertheless, instead of using terms just as means of label propagation, in this article we propose the use of the bipartite network structure to define the relevance scores of terms for classes through an optimization process and then propagate these relevance scores to define labels for unlabeled documents. The new document labels are used to redefine the relevance scores of terms which consequently redefine the labels of unlabeled documents in an iterative process. We demonstrated that the proposed approach surpasses the algorithms for transductive classification based on vector space model or networks. Moreover, we demonstrated that the proposed algorithm effectively makes use of unlabeled documents to improve classification and it is faster than other transductive algorithms.

Text clustering is a text mining task which is often used to aid the organization, knowledge extraction, and exploratory search of text collections. Nowadays, the automatic text clustering becomes essential as the volume and variety of digital text documents increase, either in social networks and the Web or inside organizations. This paper explores the use of named entities as privileged information in a hierarchical clustering process, so as to improve clusters quality and interpretation. We carried out an experimental evaluation on three text collections (one written in Portuguese and two written in English) and the results show that named entities can be applied as privileged information to power clustering solution in dynamic text collection scenarios.

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.

Usually, algorithms for categorization of numeric data have been applied for text categorization after a preprocessing phase which assigns weights for textual terms deemed as attributes. However, due to characteristics of textual data, some algorithms for data categorization are not efficient for text categorization. Characteristics of textual data such as sparsity and high dimensionality sometimes impair the quality of general purpose classifiers. Here, we propose a text classifier based on a bipartite heterogeneous network used to represent textual document collections. Such algorithm induces a classification model assigning weights to objects that represents terms of the textual document collection. The induced weights correspond to the influence of the terms in the classification of documents they appear. The least-mean-square algorithm is used in the inductive process. Empirical evaluation using a large amount of textual document collections shows that the proposed IMBHN algorithm produces significantly better results than the k-NN, C4.5, SVM and Naïve Bayes algorithms.

A simple and intuitive way to organize a huge document collection is by a topic hierarchy. Generally two steps are carried out to build a topic hierarchy automatically: 1) hierarchical document clustering and 2) cluster labeling. For both steps, a good textual document representation is essential. The bag-of-words is the common way to represent text collections. In this representation, each document is represented by a vector where each word in the document collection represents a dimension (feature). This approach has well known problems as the high dimensionality and sparsity of data. Besides, most of the concepts are composed by more than one word, as "document engineering" or "text mining". In this paper an approach called bag-of-related-words is proposed to generate features compounded by a set of related words with a dimensionality smaller than the bag-of-words. The features are extracted from each textual document of a collection using association rules. Different ways to map the document into transactions in order to allow the extraction of association rules and interest measures to prune the number of features are analyzed. To evaluate how much the proposed approach can aid the topic hierarchy building, we carried out an objective evaluation for the clustering structure, and a subjective evaluation for topic hierarchies. All the results were compared with the bag-of-words. The obtained results demonstrated that the proposed representation is better than the bag-of-words for the topic hierarchy building.

The association rule mining task identifies all the intrinsic associations among the items contained in data and leads to only specialized knowledge. To overcome this problem the generalized association rules appeared. This type of rule associates not only the items contained in data, but also some items encoded into a given taxonomy. Therefore, the techniques used to obtain generalized association rules are very useful since they provide a more general view of the domain. However, a problem found when using these techniques is how to identify the most useful rules to avoid overload the user with a huge amount of patterns. Nowadays, the researches use objective evaluation measures to evaluate and select the most interesting knowledge to the user. Despite the fact these measures have been studied by many researches to evaluate many types of rules (for example, classification and traditional association rules), it is important to study these measures in the context of generalized rules. Thus, this paper presents an analytical evaluation to understand the behavior of some objective measures when applied in a set of generalized rules. Many relations were obtained to express the behavior of these measures, what represents a meaningful contribution to the post-processing data mining area