Improving Semantic Knowledge Base for Transfer Learning in Sentiment Analysis

Transcription

1 109 Improving Semantic Knowledge Base for Transfer Learning in Sentiment Analysis R.Gayathri,1, K. Krishna Kumari 2 1 P.G Student, 2 Associate Professor Department of Computer Science and Engineering, A.V.C College of Engineering, Mayiladuthurai, Tamil Nadu Abstract Sentiment analysis deals with the computational treatment of opinion, sentiment, and subjectivity in text, has attracted a great deal of attention. Sentiment analysis has been widely used across a wide range of domains in recent years, such as information retrieval, question answering systems and social network. This paper presents a new method for improving the semantic knowledge base for sentiment classification in social web applications. It comprises the three steps. First, to identify sentiment terms. Next, to provide the context information from training corpus and ground this information to lexical resources such as WordNet. This Work applies to a transfer learning method called cross-domain sentiment classification. In Sentiment Analysis, transfer learning can be applied to transfer sentiment classification from one domain to another or building a bridge between two domains. This is achieved by learning the semantic knowledge base across the different domains. A model called AS_LDA is used for the sentiment classification. The performance of the proposed system improves the accuracy of the Sentiment Classifier to a significant extent. Key terms: WordNet, Sentiment Analysis, Cross-Domain sentiment Classification, Transfer Learning. 1. INTRODUCTION Sentiment analysis is a technique to classify people s opinions in product reviews, blogs or social networks. Large datasets are available on-line today, they can be numerical or text file and they can be structured, semistructured or non-structured. Approaches and technique to apply and extract useful information from these data have been the major focuses of many researchers and practitioners lately. Many different information retrieval techniques and tools have been proposed according to different data types. In addition to data and text mining, there has seen a growing interest in non-topical text analysis in recent years. Sentiment analysis is one of them. Sentiment analysis, also known as opinion mining, is to identify and extract subjective information in source materials, which can be positive, neutral, or negative. Sentiment Analysis allows business to track the Flame detection, new product perception, Brand Perception and Reputation Management. It allows the individuals to get opinion on something on a global Scale. Sentiment is expressed differently in dissimilar domains, and it is costly to interpret data for each new domain in which we would like to apply a sentiment classifier. Here, we proposed a cross-domain classification method that overcomes those challenges. 1.1 Problem Description Many examples in knowledge engineering can be found where transfer learning can truly be beneficial. Consider the problem of sentiment classification, where our task is to automatically classify the reviews on a product, such as a brand of camera, into positive and negative views. For this classification task, we need to first collect many reviews of the product and annotate them. We would then train a classifier on the reviews with their corresponding labels. Since the distribution of review data among different types of products can be very different, to maintain good classification performance, we need to collect a large amount of labeled data in order to train the review-classification models for each product. However, this data-labeling process can be very expensive to do. To reduce the effort for annotating reviews for various products, we may want to adapt a classification model that is trained on some products to help learn classification models for some other products. In this paper, a new Knowledge Based Transfer Learning (KBTL) model is proposed. 1.2 Sentiment Classification Sentiment classification is an opinion mining activity concerned with determining what, if any, is the Overall sentiment orientation of the opinions contained within a given document. It is assumed in general that the document being inspected contains subjective information, such as in product reviews and feedback forms. Opinion orientation can be classified as belonging to opposing positive or negative polarities positive or negative

2 110 feedback about a product, favorable or unfavorable opinions on a topic 1.3 Transfer Learning Method Transfer learning extracts knowledge from auxiliary domain to improve the learning process in a target domain. Transfer learning is considered a new cross domain learning technique as it addresses the various aspects of domain differences. In Sentiment Analysis; transfer learning can be applied to transfer sentiment classification from one domain to another or building a bridge between two domains. This proposed system accurately Transfer the sentiment classification across the different domain using the enriched knowledge base. 2. RELATED WORK Common Sentiment Analysis Task [5] proposed the basic task of opinion mining is polarity classification and Agreement detection. Polarity classification occurs when a piece of text stating an opinion on a single issue is classified as one of two opposing sentiments. Polarity classifications also identify pro and con expressions in online reviews. Agreement detection determines whether a pair of text documents should receive the same or different sentiment-related labels. WORDNET relations [2] proposed WORDNET-AFFECT, generates synsets that still represent affective concepts. If the resulting synsets are members of WORDNET- AFFECT, then the answer is trivially affirmative. For other relations such as hyperonymy, entailment, causes, verbgroup it assumed the affective mean and it is necessary to manually filter the synsets in order to select those affective concepts. NLP curves [6] proposed the automated analysis techniques for extract and manipulate text meanings. A NLP system must have access to a significant amount of knowledge about the world and the domain of discourse. Text categorization [17] addresses the use of Support Vector Machine (SVM), based on principle of Structural risk minimization it make the analysis with particular properties of learning with text data. It finds the good parameters automatically using structural risk minimization principle. Domain adaptation machine [4] proposed a robust classifier for the target domain by leveraging many base classifiers which could be learned using the labeled samples from the source domains or the target domains. 3. PROPOSED SYSTEM The proposed knowledge base can be used to fix the existing context-aware approaches use vector space address the problem of contextual polarity change. This aims to increase the lexicons coverage and derive information for subsequent sentiment analysis. We use WordNet terms and their polarity values to generate a baseline sentiment lexicon, identify sentiment terms, and extract context information from training corpus and ground this information to lexical resources such as WordNet. This knowledge base is to make as domain adaptation for cross Domain sentiment classification. It provides a two-stage framework for cross-domain sentiment classification. In the first stage they built a bridge between the source domain and the target domain to get some most confidently labeled documents in the target domain. In the second stage they exploited the intrinsic structure, revealed by these labeled documents to label the target-domain data. WordNet is a lexical database for English language that group s English word into set of synonyms called synset. WordNet distinguish between nouns, verbs, adjective as major categories. At Concept level, WordNet which is given in Figure 1 is used as a knowledge base for deriving the semantic and lexical relations. Transfer component analysis [15] proposed the methods aim to extract a shared feature subspace in which the difference in distributions across domains can be reduced by minimizing predefined distance measures. Structural correspondence learning [9] proposed the algorithm for linking the source and target domain by selecting the pivot features and highly deals with correct misalignment using labeled instances. Fig.1 WordNet

3 111 This section summarizes the sequence of steps to identify the Sentiment terms and ground them to common-sense and common knowledge, i.e. WordNet senses. The gained knowledge is used to improve concepts with context information from WordNet definitions and statements. The following Figure 2 illustrates the overall architecture of the KBTL system. Training Data Test Data Feature Extraction Preprocessor Word Splitter Removing Stop Words Domain Specific Keywords Cross Domain Sentiment Classification a. Lexical Analyzer: Converts the plain input text into an output token stream. This module is produced with the JavaCC2 parser generator. Additionally, this module spots the possible affective containers (content words), valence shifters such as negation words and intensifiers and filters out stop words like function words. b. Sentence Splitter: Delimits the sentences through a binary decision tree following. In general, periods, uppercase letters, exclamation points and question marks are good indicators of sentence boundaries. c. POS Tagger: Determines the function of nouns, verbs and adjectives (classes of words with a possible affective content) within the sentence. A statistical approach is implemented using the Stanford log-linear POS tagger. Test Data Preprocessing Lexical Analyzer POS Tagger Accuracy Detection Splitter Fig.2 Overview of the Proposed KBTL Model 3.1 Training Data Set The training data should be extracted as feature and it can be trained for making decision making with the test data from the same domain. The domain specific keywords are collected and made as input for the training as well as testing. The most important indicators of sentiments are sentiment words, also called opinion words. These are words that are commonly used to express positive or negative sentiments. Although sentiment words and phrases are important for sentiment analysis, only using them is far from sufficient. The problem is much more complex. In other words, we can say that sentiment lexicon is necessary but not sufficient for sentiment analysis. 3.2 Preprocessing the Test Data The test data should be collected and preprocessed for the process of classification. The Figure 3 shows the various involved in the preprocessing which are explained as below POS Tagger Fig.3 Preprocessing System 3.3 Feature Extraction The Classifier predicts the most appropriate sentiment label according to the features extracted from the terms observed in the text, which is usually taken for a bag of words. Word-Sense Disambiguator: resolves the meaning of affective words (i.e., nouns, adjectives and verbs) according to their context. It uses a semantic similarity measure to score the senses of an affective word with the context words using the WordNet ontology. Additionally, the module retrieves the set of synonyms for the resulting sense in order to expand the feature space. It performs a series of processing steps to ground a sentiment terms to WordNet concepts involves, extract positive and negative context terms from the contextualized sentiment lexicon 3.4 Cross Domain Sentiment Classifications Bag of Words Supervised machine learning techniques are used for classified document or sentences into finite set of class i.e into positive, negative and neutral. Training data set is

4 Actual Class International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN available for all kind of classes. An optimal scenario will allow for the algorithm to correctly determine the class labels for unseen instances. We are using the sentiment model called Auxiliary-Sentiment Latent Dirichlet Allocation (AS-LDA) for sentiment classification. It identifies the polarity of the subjective document using the sentiment element words and auxiliary words which are sampled accordingly from sentiment topics and auxiliary topics. Sentiment element words include targets of the opinions, polarity words and modifiers of polarity words. 3.5 Accuracy Detection Accuracy for this classifier can be detected using the Confusion Matrix which is given Table 1. Confusion matrix is a tool for analyzing how well classifier can recognize tuples of different classes. A confusion matrix contains information about actual and predicted classifications done by a classification system. Accuracy detects the percentage of predictions that are correct using formula (TP + TN) / (TP + TN + FP + FN). Table 1 Confusion Matrix Predicted Class YES NO YES TP FN NO FP TN 4. RESULTS AND DISCUSSION We model this classifier by the training the two different domain of Hotel, Music Datasets and tested against the laptop domain. Sentiment is expressed differently in different domains, and it is costly to annotate data for each new domain in which we would like to apply a sentiment classifier. It can be done by training a classifier from one or more source domains and applying the trained classifier on a different domain target domain. The sentiment classifications are explained in the proposed method. While classifying the dataset across the domain it achieves the higher accuracy compared to single domain classification. The corresponding Cross-domain classifiers achieve the performance of 86.15% in accuracy, respectively. Below graphical representation in Figure 4 shows the implementation of proposed sentiment classification KBTL model results compared with existing cross domain sentiment classification methods such as Structural Correspondence Learning (SCL) algorithm, Sentiment Sensitive Thesaurus (SST) algorithm. KBTL SST SCL Fig.4 Accuracy of the KBTL Classifier 5. CONCLUSION Sentiment Analysis, as an interdisplinary field that crosses natural language processing, artificial intelligence and text mining. This paper presents a new method to increase the lexicon exposure and effectively derive the concept information for sentiment classification. Improving knowledge bases with information on (i) sentiment terms, (ii) positive and negative context terms, (iii) the grounding of this information to common-sense and common knowledge bases such as WordNet. This knowledge base is used to make as domain adaptation for cross Domain sentiment classification. It can be done by transfer sentiment classification from one domain to another or building a bridge between two domains. This is achieved by learning the semantic knowledge base across the different domains. The AS_LAD classification algorithm is proposed to obtain the effective results in the sentiment classification. It evaluates the results through the supervised learning algorithms. In future, it is planned to generalize the proposed method to solve other types of domain adaptation tasks. REFERENCES [1] A. Das, B. Gambaeck, Sentimantics: conceptual spaces for lexical sentiment polarity representation with contextuality, Proceedings of 3rd Workshop on Computational Approaches to Subjectivity and Sentiment

5 113 Analysis at ACL (WASSA), Jungmun, South Korea, pp , [2] C. Strapparava, A. Valitutti, WordNet-affect: an affective extension of WordNet, Proceedings of 4th International Conference on Language Resources and Evaluation (LREC), Lisbon, Portugal, pp , [3] D. Lenat and R. Guha, Building Large Knowledge- Based Systems: Representation and Inference in the Cyc Project, Artificial Intelligence, vol. 61, pp , Elsevier, [4] Duan,L.,Xu, D.,Tsang,I.W, Domain Adaptation from Multiple Sources:a Domain-dependent Regularization Approach, IEEE Transactions on Neural Network Learning System, vol. 23(3), pp , [5] E. Cambria, B. Schuller, Y. Xia, C. Havasi, New Avenues in Opinion Mining and Sentiment Analysis, IEEE Transactions on Intelligent System, vol. 28 (2), pp.15 21, [6] E. Cambria, B. White, Jumping NLP curves: A Review of Natural Language Processing Research, IEEE Transactions on Computational Intelligent Management, vol. 9 (2), pp , [7] H. Kanayama and T. Nasukawa, Fully Automatic Lexicon Expansion for Domain-Oriented Sentiment Analysis, Proceedings of Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP 06), pp , [8] H. Takamura, T. Inui, and M. Okumura, Extracting Semantic Orientation of Words Using Spin Model, Proceedings of Annual Meeting of the Association of Computational Linguistics (ACL 05), pp , [11] Kim Soo-Min and Eduard Hovy, Determining the Sentiment of Opinions, In Proceedings of Conference on Computational Linguistics (COLING-04), Geneva, Switzerland, pp , [12] Miller G. A., Beckwith R., Fellbaum C, Gross D, Miller K. J. (1990), Introduction to Wordnet: An On- line Lexical Database, International Journal of Lexicography, vol. 3, no. 4, pp , [13] N. Kaji and M. Kitsuregawa, Building Lexicon for Sentiment Analysis from Massive Collection of HTML Documents, Proceedings of Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP 07), pp , [14] P.D. Turney, Thumbs Up or Thumbs Down? Semantic Orientation Applied to Unsupervised Classification of Reviews, Proceedings of 40th Annual Meeting on Assoc. for Computational Linguistics (ACL 02), pp , [15] S. J. Pan, I. W. Tsang, J. T. Kwok, and Q. Yang, Domain adaptation via transfer component analysis, IEEE Transactions on Neural Network, vol. 22, no. 2, pp , [16] S. Poria, A. Gelbukh, A. Hussain, N. Howard, D. Das, S. Bandyopadhyay, Enhanced SenticNet with Affective Labels for Concept-based Opinion Mining, IEEE Transactions on Intelligent System, vol. 28 (2), pp , [17] T. Joachims, Text Categorization with Support Vector Machines:Learning with Many Relevant Features, Proceedings of 10th European Conference on Machine Learning (ECML 98), pp , [9] J. Blitzer, M. Dredze, and F. Pereira, Biographies, Bollywood, Boom-Boxes and Blenders: Domain Adaptation for Sentiment Classification, Proceedings of 45th Annual Meeting of the Association of Computational Linguistics (ACL 07), pp , [10] K. Yoshida, Y. Tsuruoka, Y. Miyao, and J. Tsujii, Ambiguous Part-of-Speech Tagging for Improving Accuracy and Domain Portability of Syntactic Parsers, Proceedings of 20th International Joint Conference on Artificial Intelligence (IJCAI 07), pp , 2007.