Ali Mustafa Qamar

Similarity Learning for Nearest Neighbor Classification

In this paper, we propose an algorithm for learning a general class of similarity measures for kNN classification. This class encompasses, among others, the standard cosine measure, as well as the Dice and Jaccard coefficients. The algorithm we propose is an extension of the voted perceptron algorithm and allows one to learn different types of similarity functions (either based on diagonal, symmetric or asymmetric similarity matrices). The results we obtained show that learning similarity measures yields significant improvements on several collections, for two prediction rules: the standard kNN rule, which was our primary goal, and a symmetric version of it.

Online and Batch Learning of Generalized Cosine Similarities

In this paper, we define an online algorithm to learn the generalized cosine similarity measures for kNN classification and hence a similarity matrix A corresponding to a bilinear form. In contrary to the standard cosine measure, the normalization is itself dependent on the similarity matrix which makes it impossible to use directly the algorithms developed for learning Mahanalobis distances, based on positive, semi-definite (PSD) matrices. We follow the approach where we first find an appropriate matrix and then project it onto the cone of PSD matrices, which we have adapted to the particular form of generalized cosine similarities, and more particularly to the fact that such measures are normalized. The resulting online algorithm as well as its batch version is fast and has got better accuracy as compared with state-of-the-art methods on standard data sets.

Prediction and analysis of Pakistan election 2013 based on sentiment analysis

The significance of social media has already been proven in provoking transformation of public opinion for developed countries in improving democratic process of elections. On the contrary, developing countries lacking basic necessities of life possess monopolistic electoral system in which candidates are elected based on tribes, family backgrounds, or landlord influences. They extort voters to cast votes against their promises for the provision of basic needs. Similarly voters also poll votes for personal interests being unaware of party manifesto or national interest. These issues can be addressed by social media, resulting as ongoing process of improvement for presently adopted electoral procedures. People of Pakistan utilized social media to garner support and campaign for political parties in General Elections 2013. Political leaders, parties, and people of Pakistan disseminated partys agenda and advocacy of partys ideology on Twitter without much campaigning cost. To study effectiveness of social media inferred from individuals political behavior, large scale analysis, sentiment detection & tweet classification was done in order to classify, predict and forecast election results. The experimental results depicts that social media content can be used as an effective indicator for capturing political behaviors of different parties Positive, negative and neutral behavior of the party followers as well as partys campaign impact can be predicted from the analysis. The analytical findings proved to be having considerable correspondence with actual results as published by Election Commission of Pakistan..

A semantic rules & reasoning based approach for Diet and Exercise management for diabetics

Diabetes is a serious chronic disease and balance diet as well as regular exercise are leading important factors for diabetes control. Management of healthy diet and proper exercise involve many decision variables from different domains such as gender, weight, height, age, needed calories and nutrition values, preferences about food and exercise, clinical guidelines and current vital signs etc. We have implemented a semantic rules and reasoning based approach that generates diet and exercise recommendations for diabetes patients. The developed prototype application is named Semantic Healthcare Assistant for Diet and Exercise (SHADE). Individual ontologies are defined for different domains (person, disease, food and exercise) along with SWRL rules and then imported all into an integrated ontology. The integrated ontology semantically generates the recommendations as inferences based on data and rules by using Pellet reasoner. Each generated meal menu is a list of food items along with portion size such that food items are users preferred and menu is personalized, healthy and balanced diet. Finally, SHADE recommends users preferred activities as exercises along with duration and intensity.

Similarity Learning in Nearest Neighbor and Relief Algorithm

In this paper, we study the links between RELIEF, a well-known feature re-weighting algorithm and SiLA, a similarity learning algorithm. On one hand, SiLA is interested in directly reducing the leave-one-out error or 0-1 loss by reducing the number of mistakes on unseen examples. On the other hand, it has been shown that RELIEF could be seen as a distance learning algorithm in which a linear utility function with maximum margin was optimized. We first propose here a version of this algorithm for similarity learning, called RBS (for RELIEF-Based Similarity learning). As RELIEF, and unlike SiLA, RBS does not try to optimize the leave-one-out error or 0-1 loss, and does not perform very well in practice, as we illustrate on two UCI collections. We thus introduce a stricter version of RBS, called sRBS, aiming at relying on a cost function closer to the 0-1 loss. Experiments conducted on several datasets illustrate the different behaviors of these algorithms for learning similarities for kNN classification. The results indicate in particular that the 0-1 loss is a more appropriate cost function than the one implicitly used by RELIEF.

Classification and legality analysis of bowling action in the game of cricket

One of the hot topics in modern era of cricket is to decide whether the bowling action of a bowler is legal or not. Because of the complex bio-mechanical movement of the bowling arm, it is not possible for the on-field umpire to declare a bowling action as legal or illegal. Inertial sensors are currently being used for activity recognition in cricket for the coaching of bowlers and detecting the legality of their moves, since a well trained and legal bowling action is highly significant for the career of a cricket player. After extensive analysis and research, we present a system to detect the legality of the bowling action based on real time multidimensional physiological data obtained from the inertial sensors mounted on the bowlers arm. We propose a method to examine the movement of the bowling arm in the correct rotation order with a precise angle. The system evaluates the bowling action using various action profiles. The action profiles are used so as to simplify the complex bio-mechanical movement of the bowling arm along with minimizing the size of the data provided to the classifier. The events of interest are identified and tagged. Algorithms such as support vector machines, k-nearest neighbor, Naïve Bayes, random forest, and artificial neural network are trained over statistical features extracted from the tagged data. To accomplish the reliability of outcome measures, the technical error of measurement was adopted. The proposed method achieves very high accuracy in the correct classification of bowling action.

Refining Kea++ automatic keyphrase assignment

Keyphrases facilitate finding the right information in digital sources. Keyphrase assignment is the alignment of documents or text with keyphrases of any standard taxonomy/classification system. Kea++ is an automatic keyphrase assignment tool using a machine learning-based technique. However, it does not effectively exploit the hierarchical relations that exist in its input taxonomy and returns noise in its results. The refinement methodology was designed as a top layer of Kea++ in order to fine tune its results. It was an initial step and focused on a single Computing domain. It was neither validated on multiple domains nor evaluated to determine whether the improvement in the results is significant or not. The aim of this task was to solidify the refinement methodology. The main contributions of this work are (a) to extend the methodology for multiple domains and (b) to statistically verify that the improvement in the Kea++ results is significant.

Dynamic entity and relationship extraction from news articles

In structured as well as unstructured data, information extraction (IE) and information retrieval (IR) techniques are gaining popularity in order to produce a realistic output. The Internet users are growing day by day and becoming a popular source for spreading the information through news/blogs etc. To monitor this information, a lot of quality work has been done in that perspective. Related to news monitoring, our proposed unsupervised machine learning approach will fetch the entities and relationships from the news document itself and through comparison with other related news documents, it will form a cluster. We propose, in this paper, a dynamic model for entity extraction and relationship in order to monitor the news reported in the news articles.

Batch document filtering using nearest neighbor algorithm

We propose in this paper a batch algorithm to learn category specific thresholds in a multiclass environment where a document can belong to more than one class. The algorithm uses the k-nearest neighbor algorithm for filtering the 100,000 documents into 50 profiles. The experiments were run on the English corpus. Our experiments gave us a macro precision of 0.256 while the macro recall was 0.295. We had participated in the online task in INFILE 2008 where we had used an online algorithm using the feedbacks from the server. In comparison with INFILE 2008, the macro recall is significantly better in 2009, 0.295 vs 0.260. However the macro precision in 2008 were 0.306. Furthermore, the anticipation in 2009 was 0.43 as compared with 0.307 in 2008. We have also provided a detailed comparison between the batch and online algorithms.

TIE algorithm: a layer over clustering-based taxonomy generation for handling evolving data

Taxonomy is generated to effectively organize and access large volume of data. A taxonomy is a way of representing concepts that exist in data. It needs to continuously evolve to reflect changes in data. Existing automatic taxonomy generation techniques do not handle the evolution of data; therefore, the generated taxonomies do not truly represent the data. The evolution of data can be handled by either regenerating taxonomy from scratch, or allowing taxonomy to incrementally evolve whenever changes occur in the data. The former approach is not economical in terms of time and resources. A taxonomy incremental evolution (TIE) algorithm, as proposed, is a novel attempt to handle the data that evolve in time. It serves as a layer over an existing clustering-based taxonomy generation technique and allows an existing taxonomy to incrementally evolve. The algorithm was evaluated in research articles selected from the computing domain. It was found that the taxonomy using the algorithm that evolved with data needed considerably shorter time, and had better quality per unit time as compared to the taxonomy regenerated from scratch.