Nurfadhlina Mohd Sharef

Detecting deceptive reviews using lexical and syntactic features

Deceptive opinion classification has attracted a lot of research interest due to the rapid growth of social media users. Despite the availability of a vast number of opinion features and classification techniques, review classification still remains a challenging task. In this work we applied stylometric features, i.e. lexical and syntactic, using supervised machine learning classifiers, i.e. Support Vector Machine (SVM) with Sequential Minimal Optimization (SMO) and Naive Bayes, to detect deceptive opinion. Detecting deceptive opinion by a human reader is a difficult task because spammers try to write wise reviews, therefore it causes changes in writing style and verbal usage. Hence, considering the stylometric features help to distinguish the spammer writing style to find deceptive reviews. Experiments on an existing hotel review corpus suggest that using stylometric features is a promising approach for detecting deceptive opinions.

Text fragment extraction using incremental evolving fuzzy grammar fragments learner

Additional structure within free texts can be utilized to assist in identification of matching items and can benefit many intelligent text pattern recognition applications. This paper presents an incremental evolving fuzzy grammar (IEFG) method that focuses on the learning of underlying text fragment patterns and provides an efficient fuzzy grammar representation that exploits both syntactic and semantic properties. This notion is quantified via (i) fuzzy membership which measures the degree of membership for a text fragment in a semantic grammar class and (ii) fuzzy grammar similarity which estimates the similarity between two grammars (iii) grammar combination which combines and generalizes the grammar at a minimal generalization. Terrorism incidents data from the United States World Incidents Tracking System (WITS) are used in experiments and presented throughout the paper. A comparison with regular expression methods is made in identification of text fragments representing times. The application of text fragment extraction using IEFG is demonstrated in event type, victim type, dead count and wounded count detection with WITS XML-tagged data used as golden standard. Results have shown the efficiency and practicality of IEFG.

Order Independent Incremental Evolving Fuzzy Grammar Fragment Learner

It is generally known that most incremental learning systems are order dependent, i.e provide results that depend on the particular order of the data presentation. Our previous work has developed an incremental soft computing algorithm which can be applied to learn text fragment patterns in semi-structured texts. A set of fuzzy grammar fragments is evolved, able to recognize the string set used as examples and any similar strings. Slight modification of the grammar fragments is performed to learn new patterns. This paper investigates the theoretical aspects of order-independence in the algorithm and shows that equivalent grammar fragments are produced irrespective of the order in which illustrative examples are presented.

Review on the advancements of disambiguation in semantic question answering system

Ambiguity is a potential problem in any semantic question answering (SQA) system due to the nature of idiosyncrasy in composing natural language (NL) question and semantic resources. Thus, disambiguation of SQA systems is a field of ongoing research. Ambiguity occurs in SQA because a word or a sentence can have more than one meaning or multiple words in the same language can share the same meaning. Therefore, an SQA system needs disambiguation solutions to select the correct meaning when the linguistic triples matched with multiple KB concepts, and enumerate similar words especially when linguistic triples do not match with any KB concept. The latest development in this field is a solution for SQA systems that is able to process a complex NL question while accessing open-domain data from linked open data (LOD). The contributions in this paper include (1) formulating an SQA conceptual framework based on an in-depth study of existing SQA processes; (2) identifying the ambiguity types, specifically in English based on an interdisciplinary literature review; (3) highlighting the ambiguity types that had been resolved by the previous SQA studies; and (4) analysing the results of the existing SQA disambiguation solutions, the complexity of NL question processing, and the complexity of data retrieval from KB(s) or LOD. The results of this review demonstrated that out of thirteen types of ambiguity identified in the literature, only six types had been successfully resolved by the previous studies. Efforts to improve the disambiguation are in progress for the remaining unresolved ambiguity types to improve the accuracy of the formulated answers by the SQA system. The remaining ambiguity types are potentially resolved in the identified SQA process based on ambiguity scenarios elaborated in this paper. The results of this review also demonstrated that most existing research on SQA systems have treated the processing of the NL question complexity separate from the processing of the KB structure complexity.

Evolving fuzzy grammar for crime texts categorization

This paper introduces the evolving fuzzy grammar (EFG) method for crime texts categorization. The learning model is built based on a set of selected text fragments which are then transformed into their underlying structure called fuzzy grammars.The fuzzy notion is used because the matching, parsing and grammar derivation involves uncertainty. Fuzzy union operator is also used to combine and transform individual text fragment grammars into more general representations of the learned text fragments. The set of learned fuzzy grammars is influenced by the evolution in the seen pattern; the learned model is slightly changed (incrementally) as adaptation, which does not require the conventional redevelopment.This paper compares EFG, a novel method for learning text structure at the text fragment level Refs. [1,2] with ML in categorizing crime incidents data. In contrast, the ML methods are generally statistically founded and require the conventional train-test-test-retrain when new pattern is found.It is hypothesized in this research that this makes the whole time involved in this ML higher. An experiment is carried out to compare the performance between EFG and ML methods in precision, recall, FF-measure and extension learning. A significant test is performed to measure the difference in mean value within precision, recall and FF-measure competency.The main strength of this paper in comparison with previous related works is that it describes completely the steps involved in developing EFG. In Ref. [2] the authors presented the general step for grammar-grammar combination. In Ref. [3] the highlight was given on the step to combine grammar while the Ref. [4] emphasizes the permutation free generated fuzzy grammars.This paper however, integrates and refines the previous mentioned papers by providing more details such as the algorithm for grammar combination and demonstrates this using data that express armed attack and bombing events.This paper also ventures on a new problem on text classification compared to previous smaller scale data used in case study of EFG in Ref. [2] and text extraction task in Refs. [1,6].Results show that the EFG algorithm produces results that are close in performance with the other ML methods in terms of precision, recall, and FF-measure. The performance across approaches investigated in this paper is compared against EFG using significant test. Result has also shown that EFG has lower model retraining adaptability time. Text mining refers to the activity of identifying useful information from natural language text. This is one of the criteria practiced in automated text categorization. Machine learning (ML) based methods are the popular solution for this problem. However, the developed models typically provide low expressivity and lacking in human-understandable representation. In spite of being highly efficient, the ML based methods are established in train-test setting, and when the existing model is found insufficient, the whole processes need to be reinvented which implies train-test-retrain and is typically time consuming. Furthermore, retraining the model is not usually practical and feasible option whenever there is continuous change. This paper introduces the evolving fuzzy grammar (EFG) method for crime texts categorization. In this method, the learning model is built based on a set of selected text fragments which are then transformed into their underlying structure called fuzzy grammars. The fuzzy notion is used because the grammar matching, parsing and derivation involve uncertainty. Fuzzy union operator is also used to combine and transform individual text fragment grammars into more general representations of the learned text fragments. The set of learned fuzzy grammars is influenced by the evolution in the seen pattern; the learned model is slightly changed (incrementally) as adaptation, which does not require the conventional redevelopment. The performance of EFG in crime texts categorization is evaluated against expert-tagged real incidents summaries and compared against C4.5, support vector machines, naive Bayes, boosting, and k-nearest neighbour methods. Results show that the EFG algorithm produces results that are close in performance with the other ML methods while being highly interpretable, easily integrated into a more comprehensive grammar system and with lower model retraining adaptability time.

Sentiment Classification of Financial News Using Statistical Features

Sentiment classification of financial news deals with the identification of positive and negative news so that they can be applied in decision support systems for stock trend predictions. This paper explores several types of feature spaces as different data spaces for sentiment classification of the news article. Experiments are conducted using N-gram models unigram, bigram and the combination of unigram and bigram as feature extraction with traditional feature weighting methods (binary, term frequency (TF), and term frequency-document frequency (TF-IDF)), while document frequency (DF) was used in order to generate feature spaces with different dimensions to evaluate N-gram models and traditional feature weighting methods. We performed some experiments to measure the classification accuracy of support vector machine (SVM) with two kernel methods of Linear and Gaussian radial basis function (RBF). We concluded that feature selection and feature weighting methods can have a substantial role in sentiment classification. Furthermore, the results showed that the proposed work which combined unigram and bigram along with TF-IDF feature weighting method and optimized RBF kernel SVM produced high classification accuracy in financial news classification.

AgeTrust: A New Temporal Trust-Based Collaborative Filtering Approach

Recommender systems are useful techniques for solving the problem of information overload. Collaborative Filtering (CF) is the most successful approach for recommendation. This approach focuses on previous indicate preferences which is known for its traditional problems such as cold-start, sparsity and hacking. For solving the problem of hacking and improving the accuracy, trust-based CF methods have been proposed previously. These methods focused on trust values among the users. Nonetheless, most existing approaches use trust as a factor independent from time which we think that trust value between users is dynamic; hence it change over time. For this reason, we used friendship time and proposed a novel temporal-trust based approach called AgeTrust to measure trust value. To validate the proposed approach, we used Delicious data set and compared our approach with two other traditional trust-based approaches: traditional CF and FriendshipTrust. Result shows that our proposed approach outperforms the traditional approaches.

License plate localization using a Naïve Bayes classifier

This paper presents a probabilistic technique to localize license plates regions for cars adhering to the standard set by the Malaysian Road Transport Department. Images of the front/rear-view of cars displaying their license plates are firstly preprocessed, followed by features extraction generated from connected components analysis. These features are then used to train a Naïve Bayes classifier for the final task of license plates localization. Experimental results conducted on 144 images have shown that considering two candidates with the highest posterior probabilities better guarantees license plates regions are properly localized, with a recall of 0.98.

Semantic question answering of umrah pilgrims to enable self-guided education

Umrah as a pilgrimage to Mecca is obligatory in Islam madhab. Pilgrims can gain their knowledge about the requirements of Umrah using books, expert and existent question answering systems but still they suffer from the lack of complexity and natural language patterns. This paper proposes the semantic based question answering system to able pilgrims to compose any question about Umrah in natural language format. The proposed system used ontology to represent the knowledge about ritual of Umrah and Pilgrims. The question complexity for question in natural language is observed since it needs to be map with the contents in the ontology. An Umrah Knowledge module in this system covers all the rules and fact in the ontology format and the Educational modules is responsible for question answering interaction.

A temporal-focused trustworthiness to enhance trust-based recommender systems

Collaborative Filtering (CF) is the most successful technology for recommender systems. The technology does not rely on actual content of the items, but instead requires users to indicate preferences, most commonly in the form of ratings. While CF is known for its traditional problems such as cold-start, sparsity and modest accuracy, a trust-based CF has been previously proposed to solve such issues by focusing on trust values among the users. Nonetheless, all existing trust-based approaches use trust as a factor independent from scope, whether explicit or implicit. We argue that trustworthiness should not be the same across all conditions; hence the trust values should change to suit certain scope or focused area. To validate the proposed temporal-focused trustworthiness in this paper, we propose a novel pheromone-based approach to calculate trustworthiness by focusing on time factor. Implementation of the proposed approach is hoped to reduce cold-start and sparsity as well as improve accuracy of the recommendation results.