International Journal of Computer Engineering and Applications, Volume XII, Issue IV, April 18, ISSN

Transcription

1 International Journal of Computer Engineering and Applications, Volume XII, Issue IV, April 18, ISSN Furqan Iqbal Department of Computer Science and Engineering, Lovely Professional University, Phagwara, Punjab ABSTRACT: Nowadays Online Social Networks are so popular that they are become a major component of an individual s social interaction. They are also emotionally-rich environments where users share their emotions, feelings, ideas and thoughts. In this paper, a novel framework is proposed for characterizing emotional interactions in social networks. The aim is to extract the emotional content of texts in online social networks. The interest is in to determine whether the text is an expression of the writer s emotions or not if yes then what type of emotion likes happy, sad, angry, disgust, fear, surprise. For this purpose, text mining techniques are performed on comments/messages from a social network. The framework provides a model for data collection, feature generation, data preprocessing and data mining steps. This paper proposes a technique using Gini Index based feature selection and ensemble learning for sentiment analysis prediction. Results are evaluated on the basis of Accuracy, Precision, Recall, Fmeasure. Keywords: Query Sentiment Analysis, Emotion Mining, Social Media, Text Mining, Gini Index, Ensemble Learners [1] INTRODUCTION Nowadays, Social media is becoming increasingly popular considering that mobile contraptions can entry social community conveniently from anyplace. Accordingly, Social media is fitting an major topic for research in many fields. As quantity of individuals utilizing social network are developing day-to-day, to be in contact with their peers in order that they can share their private feeling every day and views are created on significant scale. Sentiment analysis refers to the use of natural language processing to identify and extract Furqan Iqbal 254

2 one-sided knowledge in source substances or with ease it refers back to the procedure of detecting the polarity of the text. It additionally referred as opinion mining, as it derives the opinion, or the angle of a user. A common procedure of utilizing that is described how humans think a few particular topics. Sentiment evaluation helps in deciding upon the thoughts of a speaker or a writer with respect to a few discipline topics or the total contextual polarity of a record. The perspective is also his or her choice or estimate, the emotional state of the user at the same time writing. Sentiment analysis can be used to investigate sentiment on a style of stage. It's going to ranking the whole report as optimistic or negative, and it's going to additionally ranking the response of person phrases or phrases in the record. Sentiment analysis can monitor a precise topic, many organizations use it to track or notice their merchandise, services or popularity commonly. For illustration, if someone is attacking your manufacturer on social media, sentiment evaluation will be rating the post as tremendously bad, and you could create signals for posts with hyper-negative sentiment scores. Sentiment evaluation is rough. Sentiment analysis can be categorized into four main groups: keyword spotting, lexical affinity, statistical methods, and concept-level techniques. 1. Keyword spotting: In this classifying text based on the category presence of unambiguous words such as joyful, sad, scared, and tired of something. 2. Lexical affinity: It helps in detecting observable affecting words, but it also assigns subjective words a likely affinity for particular emotions or sentiments. 3. Statistical methods leverage: It is based on elements of machine learning that includes latent semantic analysis and "bag of words". 4. Concept-level leverage: It is based on essentials that consider knowledge representation forms such as ontologies and semantic networks and therefore are used for detecting semantics which provides meaning. [4] Different levels for sentiment are as follows [9][10]: Figure 1: Level of Sentiment Analysis 1. Word level: It include following steps: a) Identifying and extracting various object features that have been given by opinion holder for e.g., reviewer. b) It determines whether the opinions on the particular feature of the object are positive, negative or neutral. c) Grouping the same features: -Producing a summary of opinion of feature based on multiple reviews. 2. Sentence level: It includes following steps: Furqan Iqbal 255

3 International Journal of Computer Engineering and Applications, Volume XII, Issue IV, April 18, ISSN a) Identify subjective or opinionated sentences b) Different Classes may be objective and subjective. c) Sentiment classification on each sentence. d) Different classes may be: positive, negative and neutral. e) Assuming a sentence may contain only one opinion which may not be true in every case. f) We can also consider clauses or phrases. 3. Document or review level: It includes: a) Sentiment classification of reviews on particular object. b) Different classes are positive, negative, and neutral c) Assuming that each document or review mainly focuses on a particular object and contains opinion from a particular holder only. [2] LITERATURE REVIEW Tripathy et al. [1] carried out a examine on Classification of Sentiment Reviews using N- gram Machine Learning Approach. This paper proposes the use of 4 algorithms such as Naive Bayes (NB), Maximum Entropy (ME), Stochastic Gradient Descent (SGD), and Support Vector Machine (SVM). They have been considered for classification of human sentiments. The accuracy of various strategies are examined in order that their overall performance on the premise of parameters like f-degree do not forget accuracy and precision is accessed. Manek et al. [2] performed a look at on Box-office Forecasting primarily based on Sentiments of Movie Reviews and Independent Subspace Method.This paper proposes a way the usage of weight by using Gini Index technique for function selection and use of guide vector machines for sentiment analysis is used for prediction the usage of numerous massive movie statistics set is used. This consequences of the use of the gini index primarily based approach indicates better performance in phrases of accuracy and blunders charge. Prerna et al. [3] have implemented a device for sentiment evaluation with the aid of combining a Rule-primarily based Classifier with Supervised Learning. The rule-primarily based classifier is based totally on rules which can be depending on the occurrences of emoticons and opinion phrases in tweets. Whereas, the Support Vector Machine (SVM) is educated on semantic, dependency, and sentiment lexicon based capabilities. The tweets are categorized as fine, terrible or unknown by using the rule of thumb-primarily based classifier, and as fine, negative or neutral by means of the SVM. Batrinca et al. [4] stated an outline of software device for social media, blogs, chats, newsfeeds etc. And the way to use them for scraping, cleansing and analyzing. For scraping the social media it suggests the challenges consisting of Data cleansing, Data protection, Data analysis and Visualization and analytics Dashboard. This paper offers a survey on technique of social media, facts, carriers and analytics techniques inclusive of circulate processing, sentimental analysis. An evaluation of various tools wished for social evaluation purpose is also supplied. There has been smooth availability of APIs furnished with the aid of Twitter, Facebook and News services which brought about explosion of facts offerings for the purpose of scraping and sentiment evaluation. Furqan Iqbal 256

4 Mihanovic et al. [5] examine sentiment analysison various devices in two exclusive bureaucracy i.e. on line review and Tweets. For dictionary making, it makes use of Knime tool in each forms. Online opinions had been crawled the usage of Apache Nutch crawler even as tweets were amassed the use of Java bundle. As tweets are shorter so, variety of tweets series might be more evaluate to on line reviews. Both tweets and on-line evaluation are saved in HBase table on Apache Hadoop server. Data sets for on-line review are labeled based totally on key, PID, Review Date, Review Text, Keyword, Language at the same time as Tweets are having attribute inclusive of Key, UserScreenName, Creation Date, Text, Keyword, Language. This facts is loaded into Knime. HarunaIsah et al. [6] represents a framework to gather and examine the view of users of drug and product by way of the use of textual content mining, sentiment analysis and machine gaining knowledge of. Proposed framework for processing view of consumer on popular logo of drugs and beauty product include: (a) Text series and cleansing, on this an API call for authenticating and extracting is invoked on Facebook Graph and Twitter APIs. Twitter API consists of REST and streaming APIs used to go looking and fetch tweets. Facebook Graph API used to fetch pages, replace repute and commenting on user revel in. (b) Preprocessing phase- adapted bag of phrases representation technique. It consists of the steps: eliminate delimiters, convert all words to lower case, eliminate numbers and prevent phrases. (c) Sentiment evaluation section-it includes two tactics. (d) Evaluation segment-contingency tables and fact tables used to represent the output of classifier. Two case studies have been protected: Sentiment evaluation on Facebook remarks and Text mining and sentiment analysis of Twitter facts. Nigam et al. [7] carried out a examine on Using Maximum Entropy for Text Classification. This paper proposes using maximum entropy techniques for type of text. Maximum entropy is used for category of textual content by means of estimating the conditional distribution of the magnificence variable given the textual file. In experiments on numerous text datasets the accuracy was in comparison to naive Bayes and show that the most entropy is on occasion notably better, but additionally occasionally worse. Catal et al. [8] conducted a have a look at on A Sentiment Classification Model Based on Multiple Classifiers. In this paper vote algorithm is utilized by combining the 3 classifiers naïve bayes, help vector machines and bagging. The use of ensemble leaners approach is used right here. Experimental results show that the use of a couple of classifiers improves the overall performance of individual classifiers on a Turkish sentiment class dataset and meta classifiers. [3] PROPOSED TECHNIQUE The proposed Gini index feature selection addresses the issues of uneven distribution of prior class probability and global goodness of a feature in two stages. First, it transforms the samples space into a feature specific normalized samples space without compromising the intra-class feature distribution. In the second stage of the framework, it identifies the features that discriminates the classes most by applying gini coefficient of inequality. 1. Pre-Processing: Furqan Iqbal 257

5 International Journal of Computer Engineering and Applications, Volume XII, Issue IV, April 18, ISSN Collection of raw data and then apply filtering techniques to make that raw data into structured format. For doing the classification, Text pre-processing and feature extraction is a preliminary phase. Pre-processing involves 3 steps: a) Word parsing and tokenization: In this phase, each user review splits into words of any natural processing language. As movie review contains block of character which are referred to as token. b) Removal of stop words: Stop words are the words that contain little information so needed to be removed. As by removing them, performance increases. Here, we made a list of around 320 words and created a text file for it. So, at the time of pre-processing we have concluded this stop word so all the words are removed from our dataset i.e. filtered. c) Stemming: It is defined as a process to reduce the derived words to their original word stem. For example, talked, talking, talks as based on the root word talk. We have used Snowball stemmer to reduce the derived word to their origin. 2. Gini Index based Feature Selection: The specific algorithm: Suppose the collection of data samples is S of s having m different values of class label attribute which defines different classes of, (i = 1;2;...;m). According to the class label attribute value, S can be divided into m subsets (, i = 1; 2;... ;m). If Si is the subset of samples which belongs to class, and is the number of the samples in the subset Si, then the Gini Index of set S is Where Pi is the probability of any sample of, which is estimated by. When the minimum of GiniIndex (S) is 0, i.e. all records belong to the same category at this collection, it indicates that the maximum useful information can be obtained. When all the samples in this collection have uniform distribution for a certain category, GiniIndex(S) reaches maximum, indicating the minimum useful information obtained. The initial form of the Gini Index is used to count the impurity of attribute for classification. The smaller its value, i.e. the lesser the impurity, the better the attribute. On the other hand, measuring the purity of attribute for categorization, the bigger its value, i.e. the better the purity, the better the attribute. 3. Classification: In order to improve the performance of individual classifiers used in the paper is to the use of meta classifiers such as bagging. While the advantage of using ensemble, methods is the improvement of the performance, the disadvantage is about the time it takes to finish the training phase. However, the main concern was to build a model which has a better performance compared to the individual classifiers. This model is novel because not only it Furqan Iqbal 258

6 uses ensemble method (Voting), but also it applies a meta classifier (Bagging) as one of its classifier component. Also, parameter optimization approach was used on its individual classifier. Each component in our model learns some parts of the classification problem and we combine these hypotheses to decide the probability level. [4] CONCLUSION This paper proposes a technique using Gini Index based feature selection and ensemble learning for sentiment analysis prediction. The proposed Gini index feature selection addresses the issues of uneven distribution of prior class probability and global goodness of a feature in two stages. First, it transforms the samples space into a feature specific normalized samples space without compromising the intra-class feature distribution. In the second stage of the framework, it identifies the features that discriminates the classes most by applying gini coefficient of inequality. Movies review are collected from IMDb movies reviews repository. Various parameters are used to evaluate the results of the proposed technique. REFERENCES [1] Abinash Tripathy, Ankit Agrawal, Santanu Kumar Rath, Classification of Sentiment Reviews using N-gram Machine Learning Approach. [2] Asha S Manek, P Deepa Shenoy,M Chandra Mohan and Venugopal K R, Aspect term extraction for sentiment analysis in large movie reviews using Gini Index feature selection method and SVM classifier, Springer, pg , Feb 04,2016. [3] Prerna Chikersal, Soujanya Poria, Erik Cambria (2015). SeNTU: Sentiment Analysis of Tweets by Combining a Rule-based Classifier with Supervised Learning, Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval), [4] BogdonBatrinca, Philip C. Treleaven (2014) Social media analytics: a survey of techniques, tools and platform,department of Computer Science, Gower Street, London,UK published in Springer. [5] Ana Mihanovic, HrvojeGabelica, ZivkoKrstic (2014) Big Data and Sentiment Analysis using Knime: Online Reviews Vs. Social Media, MIPRO Opatija, Croatia [6] HarunaIsah, Paul Trundle, Daniel Neagu. (2014) Social media Analysis for product Safety using Text mining and Sentiment Analysis, Artificial Intelligence Research Group, University of Bradford, UK, IEEE. [7] Kamal Nigam, John Lafferty, Andrew McCallum, Using Maximum Entropy for Text Classification. [8] Cagatay CATAL, Mehmet NANGIR, A Sentiment Classification Model Based on Multiple Classifiers. [9] Mrs. R.Nithya, Dr. D.Maheshwari. (2014) Sentiment Analysis on Unstructured Review, International Conference on Intelligent Computing Application, IEEE, pp , March [10] Lukasz Augustyaniak, Tomasz Kajdanowicz, PrzemyslawKazienko, MarcinKulisiewicz, WlodzimierzTuliglowicz, An Approach to Sentiment Analysis of Movie Reviews: Lexicon Based vs. Classification, Springer, Vol. 8480, pp , Furqan Iqbal 259