Search

Recently, deep Convolutional Neural Network (CNN) model has achieved remarkable results in Natural Language Processing (NLP) tasks, such as information retrieval, relation classification, semantic parsing, sentence modeling and other traditional NLP tasks, etc. On the other hand, topic modeling method has been proved to be effective by exploiting hidden knowledge in a corpus of documents. Motivated from these successes, we propose a framework that takes the advantages of closure domain measure to get enriched knowledge from close domains to the training dataset to improve the CNN model, and apply a Targeted Topic Model to take more detailed exploration on each labeled aspect of an opinion. Experimental results on different scenarios show the effectiveness of the proposed framework for multi-label classification task in comparison to other related models on the same Hotel review dataset.

In process mining, trace representation has a significant effect on the process discovery problem. The challenge is to get highly informative but low-dimensional vector space from event logs. This is required to improve the quality of the trace clustering problem for generating the process models clear enough to inspect. Though traditional trace representation methods have specific advantages, their vector space often has a big number of dimensions. In this paper, we address this problem by proposing a new trace representation method based on the deep neural networks. Experimental results prove our proposal not only is better than the alternatives, but also significantly helps to reduce the dimension of trace representation.

Inconsistencies may naturally occur in the considered application domains in Artificial Intelligence, for example as a result of data mining works in distributed sources. In order to solve inconsistent knowledge, several paraconsistent description logics have been proposed. In this paper, we face the problem of concept learning for an inconsistent knowledge base system based on bisimulation. This algorithm allows learning a concept from a training information system in a paraconsistent descriptive logic system with a set of positive items, negative items, and inconsistent items. Here, we present a system for learning concept in an inconsistent knowledge base and discuss preliminary experimental results obtained in the electronic application domain.

There have been an increasing number of various machine learning-based models successfully proposed and applied for automatic chemical-induced disease (CID) relation extraction. They, however, usually require carefully handcrafted rich feature sets, which rely on expert knowledge, thus require expensive human labor but normally still cannot generalize data well enough. In this paper, we propose a CID relation extraction model that learns features automatically through a Convolutional Neural Network (CNN) instead of traditional handcrafted features. We exploit the shortest dependency path between a disease and a chemical for identifying their CID relation. Dependency relations, with and without their direction information, are further investigated. Experimental results on benchmark datasets (namely the BioCreative V dataset) are very potential, demonstrating the effectiveness of our proposed model for CID relation extraction.

We introduce comparisons with respect to information between interpretations in paraconsistent description logics and use them to define bisimilarity for such logics. As bisimilarity is a natural notion for characterizing indiscernibility in modal and description logics, it is useful for concept learning in description logics also when inconsistencies occur. We give preservation results and the Hennessy-Milner property for comparisons with respect to information in paraconsistent description logics. As consequences, we also obtain invariance results and the Hennessy-Milner property for bisimilarity in paraconsistent description logics.

This paper proposes a model which utilizes Support Vector Machines (SVMs) - a machine learning approach for recognizing textual entailment in Vietnamese text, including three steps: (1) feature extraction, (2) training and (3) judgement by voting. In the first step, many features (e.g., Euclidean distance, Cosine, if-idf, etc) were extracted to train three classification models for the second step. The final step judged whether there is an entailment relation between a text and a hypothesis (another text can be plausibly inferred from the original one) or not. To improve the recognition quality, a combination of classifiers was proposed under voting method as human judgement on Vietnamese version of RTE-3. By using voting, our approach obtained significant improvements (from 1.2% to 9.4% of F-score) in comparison with baselines and ensemble methods, e.g. AdaBoost and Bagging.

As the number of users using Twitter1 increases, an user may have a lot of friends whose tweet (posting) list (also called as “social update stream” [5, 8, 18]) may overwhelm his/her homepage. This can lead to the situation where important tweets (i.e. the tweets the user is interested in) are pushed down on the list, thus, it takes time to find them. Social update stream ranking is a possible solution that puts important tweets on the top of the page, so that the user can easily read it. In this paper, we propose to apply hidden topics [1, 15, 20] in the Combined Regression Ranking algorithm [2] to rank social update streams. The proposed system works like a content based recommendation system. The experimental results show a significant improvement proving that our proposal is a suitable direction.

Named Entity Recognition (NER) is an important, useful task in many natural language processing applications and much previous work in NER has been done in many other languages such as English, Japanese, Chinese However, Vietnamese NER task is still relatively new and challenge due to the characteristics of Vietnamese, the lack of a large annotated corpus This paper presents a new approach for Vietnamese NER -- a semi-supervised training method for Conditional random fields (CRFs) models using generalized expectation criteria to express a preference for parameter settings. We perform several experiments using different feature setting and different training data to show the high performance of this method and compare to the other method.

The work [1] by Nguyen and Szalas is a pioneering one that uses bisimulation for machine learning in the context of description logics. In this paper we generalize and extend their concept learning method [1] for description logic-based information systems. We take attributes as basic elements of the language. Each attribute may be discrete or numeric. A Boolean attribute is treated as a concept name. This approach is more general and much more suitable for practical information systems based on description logic than the one of [1]. As further extensions we allow also data roles and the concept constructors "functionality" and "unquantified number restrictions". We formulate and prove an important theorem on basic selectors. We also provide new examples to illustrate our approach.

Feature-based opinion mining is an interesting opinion mining issue. For this problem, feature words/phrases are discovered at sentence level. However, customers usually use different words/phrases referring to the same feature in reviews. To produce a meaningful summary, synonym feature words/phrases in domain, need to be grouped under the same feature. This paper proposes a solution for grouping synonym features in Vietnamese customer reviews based on semi-supervised SVM-kNN classification and HAC clustering. Experimental results on reviews in mobile phone domain demonstrate that the proposed method is promising for the task. The Purity, Accuracy measures are 0.68 and 0.65 respectively.

Co-reference resolution task still poses many challenges due to the complexity of the Vietnamese language, and the lack of standard Vietnamese linguistic resources. Based on the mention-pair model of Rahman and Ng. (2009) and the characteristics of Vietnamese, this paper proposes a model using support vector machines (SVM) to solve the co-reference in Vietnamese documents. The corpus used in experiments to evaluate the proposed model was constructed from 200 articles in cultural and social categories from vnexpress.net newspaper website. The results of the initial experiments of the proposed model achieved 76.51% accuracy in comparison with that of the baseline model of 73.79% with similar features.

Personal names are among one of the most frequently searched items in web search engines and a person entity is always associated with numerous properties. In this paper, we propose an integrated model to recognize person entity and extract relevant values of a pre-defined set of properties related to this person simultaneously for Vietnamese. We also design a rich feature set by using various kind of knowledge resources and a apply famous machine learning method CRFs to improve the results. The obtained results show that our method is suitable for Vietnamese with the average result is 84 % of precision, 82.56% of recall and 83.39 % of F-measure. Moreover, performance time is pretty good, and the results also show the effectiveness of our feature set.

The Web contains an enormous amount of information which is formatted for human beings. This makes it difficult for computer to extract relevant content from various sources. This paper presents an XPath-wrapper induction algorithm which leverages user queries and template-based sites for extracting structured information. Our experiments show average accuracy of 94%.

In online contextual advertising, ad messages are displayed related to the content of the target Web page. It leads to the problem in information retrieval community: how to select the most relevant ad messages given the content of a page. To deal with this problem, we propose a framework that takes advantage of large scale external datasets. This framework provides a mechanism to discover the semantic relations between Web pages and ad messages by analyzing topics for them. This helps overcome the problem of mismatch due to unimportant words and the difference in vocabularies between Web pages and ad messages. The framework has been evaluated through a number of experiments. It shows a significant improvement in accuracy over word/lexicon-based matching and ranking methods.