Akshi Kumar

ComEx Miner: Expert Mining in Virtual Communities

The utilization of Web 2.0 as a platform to comprehend the arduous task of expert identification is an upcoming trend. An open problem is to assess the level of expertise objectively in the web 2.0 communities formed. We propose the “ComEx Miner System” that realizes Expert Mining in Virtual Communities, as a solution for this by quantifying the degree of agreement between the sentiment of blog and respective comments received and finally ranking the blogs with the intention to mine the expert, the one with the highest rank score. In the proposed paradigm, it is the conformity & proximity of sentimental orientation of community member’s blog & comments received on it, which is used to rank the blogs and mine the expert on the basis of the blog ranks evaluated. The effectiveness of the paradigm is demonstrated giving a partial view of the phenomenon. The initial results show that it is a motivating technique.

Alleviating Sparsity and Scalability Issues in Collaborative Filtering Based Recommender Systems

Commercial recommender systems in general are used to evaluate very large product sets. In a user – item rating database, though users are very active, there are a few rating of the total number of items available. The user-item matrix is thus extremely sparse. Since a collaborative filtering algorithm is mainly based on similarity measures computed over the co-rated set of items, the large levels of sparsity can lead to less accuracy and can challenge the predictions or recommendations of the collaborative filtering (CF)systems. Further, a CF algorithm is assumed to be efficient if it is able to filter items that are interesting to users. But, they require computations that are very expensive and grow non-linearly with the number of users and items in a database. In general, the whole ratings database is searched in collaborative filtering and thus it suffers from poor scalability when more and more users and items are added into the database. Instigated by these challenges, we investigate two collaborative filtering algorithms, firstly an algorithm based on weighted slope one scheme and item clustering & secondly an algorithm based on item classification & item clustering, which deal with the sparsity and scalability issues simultaneously. Experiments were carried to determine which is better in terms of simplicity and accuracy among the two methods.

Community Expert based Recommendation for solving First Rater Problem

ABSTRACT Information overload on the Web is a well recognized problem [1], where users find it increasingly difficult to locate the right information at the right time. Recommender system [2, 3] comes to the rescue for such a consumer. However, despite all advances, the current generation of recommender systems still requires further improvements to make recommendation methods more effective and applicable to an even broader range of real-life applications. We propose and investigate CER system, a Community Expert based Recommendation system. The paradigm is realized by an Interest Mining module which defines a constructing algorithm for Interest Group by uncovering shared interest relationships between people, using their blog document entries and interest similarity relations. Once the interest similarity group is constructed, then we identify an expert from each of the groups so formed. Expert identification from the Collaborative Interest Group is the key to recommendation as it is only the expert‟s blog whose recommendation is considered compared to systems which require a large set of customer preferences for predicting the new preferences accurately for effective Collaborative filtering-based recommendation, solving the most prominent problem existent in collaborative filtering, the

Emotion analysis of Twitter using opinion mining

With the rise in use of micro-blogging sites like Twitter, people are able to express and share their opinions with each other on a common platform. Currently all work in opinion mining research has quantified & assessed the expression of opinion as positive, negative or neutral values, we intend to categorize the opinion on the basis of five emotions, namely Happiness, Anger, Fear, Sadness & Disgust, which have been globally accepted & defined in human psychology. This paper presents a method to assess these identified types of emotions in a tweet using opinion mining. A two-step approach is proposed, where firstly, to identify the sentiment; we extract the opinion words (a combination of the adjectives along with the verbs and adverbs) in the tweets and subsequently use a novel algorithm to find the emotion values of opinion words. The initial results show that it is a motivating technique, which may find potential applications in business intelligence, government policy making, amongst others.

An Algorithmic Framework for Collaborative Interest Group Construction

As organizations, both business and research-development continue to search better ways to exploit knowledge capital accumulated on the diversified Web; it fosters the need of collaboration among people with similar interest & expertise. In this paper we focus on the problem of discovering people who have particular interests or expertise. The standard approach is to build interest group lists from explicitly registered data. However, doing so assumes one knows what lists should be built, and who ought to be included in each list. We present an alternative approach, which can support a finer grained and dynamically adaptive notion of shared interests. Our approach deduces shared interest relationships between people based on interest similarity calculated by the means of entries written on their blog. Using this approach, a user could search for people by requesting a list of people whose interests are similar to several people known to have the interest in question.

Firefly Algorithm for Feature Selection in Sentiment Analysis

Selecting and extracting feature is a vital step in sentiment analysis. The statistical techniques of feature selection like document frequency thresholding produce sub-optimal feature subset because of the non-polynomial (NP)-hard character of the problem. Swarm intelligence algorithms are used extensively in optimization problems. Swarm optimization renders feature subset selection by improving the classification accuracy and reducing the computational complexity and feature set size. In this work, we propose firefly algorithm for feature subset selection optimization. SVM classifier is used for the classification task. Four different datasets are used for the classification of which two are in Hindi and two in English. The proposed method is compared with feature selection using genetic algorithm. This method, therefore, is successful in optimizing the feature set and improving the performance of the system in terms of accuracy.

SentIndiGov-O: an Ontology-based Tool for Sentiment Analysis to Empower Digital Governance

Social media can be used to facilitate interaction between people and can offer a substantial, unparalleled platform for extensive involvement of citizens. We propose an ontology-based analytics on social media that expounds an intelligent governance model where sentiment can be mined for extracting views of citizens towards government practices, policies, rules and monitoring performance.

Image Sentiment Analysis Using Convolutional Neural Network

Visual media is one of the most powerful channel for expressing emotions and sentiments. Social media users are gradually using multimedia like images, videos etc. for expressing their opinions, views and experiences. Sentiment analysis of this vast user generated visual content can aid in better and improved extraction of user sentiments. This motivated us to focus on determining ‘image sentiment analyses’. Significant advancement has been made in this area, however, there is lot more to focus on visual sentiment analysis using deep learning techniques. In our study, we aim to design a visual sentiment framework using a convolutional neural network. For experimentation, we employ the use of Flickr images for training purposes and Twitter images for testing purposes. The results depict that the proposed ‘visual sentiment framework using convolutional neural network’ shows improved performance for analyzing the sentiments associated with the images.

Optimal test sequence generation using River Formation Dynamics

Software testing is a complex and exhaustive process, often limited by the resources. Although many approaches for test sequence generation exist in the literature, but none of it is ideal as far as coverage and redundancy is concerned. This paper aims at improving the efficiency of software testing process by generating the optimal test sequences in the control flow graph (CFG) of the program under test (PUT) by using a novel swarm intelligence method called River Formation Dynamics(RFD). RFD is inspired by a natural phenomenon of how drops transformed into river and river into sea. It provides full path coverage with zero edge/transition redundancy. It also tries to prioritize the paths based on their strength, calculated in terms of their traversal by the drops.

Sentiment analysis using neural network

With the rapid growth in use of social networking sites in the past decade, it has become a notable medium for people to express their views or opinions. This has fostered & promoted sentiment analysis as a dynamic & potential area of research where new techniques & models need to be explored for continuous improvement in result accuracy. In this paper, we propose a probabilistic neural network (PNN) with a self-adaptive approach to perform sentiment analysis on tweets. Probabilistic Neural Network as a multi-layered feed-forward neural network is an apt choice because of its prominent features of adaptive learning, fault tolerance, parallelism and generalization which provide a superior performance. Also, the smoothing parameter of PNN plays a great role for predicting an accurate class of classifier. So a self-adaptive algorithm is used to calculate and optimize the smoothing parameter in our research. Two types of Probabilistic Neural Network models are implemented in the proposed approach. First model of PNN, also called as PNNS has single value of smoothing parameter for whole network. Second model, also called as PNNC has different values of smoothing parameter for each class. The training and testing dataset is collected from Twitter using Twitter API. Accuracy of both model PNNS and PNNC is calculated and result shows that the PNNC has a better performance than PNNS.