Soujanya Poria

Enhanced SenticNet with Affective Labels for Concept-Based Opinion Mining

SenticNet 1.0 is one of the most widely used, publicly available resources for concept-based opinion mining. The presented methodology enriches SenticNet concepts with affective information by assigning an emotion label.

Enriching SenticNet Polarity Scores through Semi-Supervised Fuzzy Clustering

SenticNet 1.0 is one of the most widely used freely-available resources for concept-level opinion mining, containing about 5,700 common sense concepts and their corresponding polarity scores. Specific affective information associated to such concepts, however, is often desirable for tasks such as emotion recognition. In this work, we propose a method for assigning emotion labels to SenticNet concepts based on a semi-supervised classifier trained on WordNet-Affect emotion lists with features extracted from various lexical resources.

Fuzzy clustering for semi-supervised learning --- case study: construction of an emotion lexicon

We consider the task of semi-supervised classification: extending category labels from a small dataset of labeled examples to a much larger set. We show that, at least on our case study task, unsupervised fuzzy clustering of the unlabeled examples helps in obtaining the hard clusters. Namely, we used the membership values obtained with fuzzy clustering as additional features for hard clustering. We also used these membership values to reduce the confusion set for the hard clustering. As a case study, we use applied the proposed method to the task of constructing a large emotion lexicon by extending the emotion labels from the WordNet Affect lexicon using various features of words. Some of the features were extracted from the emotional statements of the freely available ISEAR dataset; other features were WordNet distance and the similarity measured via the polarity scores in the SenticNet resource. The proposed method classified words by emotion labels with high accuracy.

Semantic Textual Entailment Recognition using UNL

A two-way textual entailment (TE) recognition system that uses semantic features has been described in this paper. We have used the Universal Networking Language (UNL) to identify the semantic features. UNL has all the components of a natural language. The development of a UNL based textual entailment system that compares the UNL relations in both the text and the hypothesis has been reported. The semantic TE system has been developed using the RTE-3 test annotated set as a development set (includes 800 text-hypothesis pairs). Evaluation scores obtained on the RTE-4 test set (includes 1000 text-hypothesis pairs) show 55.89% precision and 65.40% recall for YES decisions and 66.50% precision and 55.20% recall for NO decisions and overall 60.3% precision and 60.3% recall.

Music Genre Classification: A Semi-supervised Approach

Music genres can be seen as categorical descriptions used to classify music basing on various characteristics such as instrumentation, pitch, rhythmic structure, and harmonic contents. Automatic music genre classification is important for music retrieval in large music collections on the web. We build a classifier that learns from very few labeled examples plus a large quantity of unlabeled data, and show that our methodology outperforms existing supervised and unsupervised approaches. We also identify salient features useful for music genre classification. We achieve 97.1% accuracy of 10-way classification on real-world audio collections.

A classifier based approach to emotion lexicon construction

The present task of developing an emotion lexicon shows the differences from the existing solutions by considering the definite as well as fuzzy connotation of the emotional words into account. A weighted lexical network has been developed on the freely available ISEAR dataset using the co-occurrence threshold. Two methods were applied on the network, a supervised method that predicts the definite emotion orientations of the words which received close or equal membership values from the first method, Fuzzy c-means clustering. The kernel functions of the two methods were modified based on the similarity based edge weights, Point wise Mutual Information (PMI) and universal Law of Gravitation (LGr) between the word pairs. The system achieves the accuracy of 85.92% in identifying emotion orientations of the words from the WordNet Affect based lexical network.

Sentic Computing for Social Media Analysis, Representation, and Retrieval

As the web is rapidly evolving, web users are evolving with it. In the era of social colonisation, people are getting more and more enthusiastic about interacting, sharing and collaborating through social networks, online communities, blogs, wikis and other online collaborative media. In recent years, this collective intelligence has spread to many different areas in the web, with particular focus on fields related to our everyday life such as commerce, tourism, education, and health. These online social data, however, remain hardly accessible to computers, as they are specifically meant for human consumption. To overcome such obstacle, we need to explore more concept-level approaches that rely more on the implicit semantic texture of natural language, rather than its explicit syntactic structure. To this end, we further develop and apply sentic computing tools and techniques to the development of a novel unified framework for social media analysis, representation and retrieval. The proposed system extracts semantics from natural language text by applying graph mining and multidimensionality reduction techniques on an affective common sense knowledge base and makes use of them for inferring the cognitive and affective information associated with social media.

SMSFR: SMS-Based FAQ retrieval system

The paper describes an SMS-based FAQ retrieval system. The goal of this task is to find a question Q* from corpora of FAQs (Frequently Asked Questions) that best answers or matches the SMS query S. The test corpus used in this paper contained FAQs in three languages: English, Hindi and Malayalam. The FAQs were from several domains, including railway enquiry, telecom, health and banking. We first checked the SMS using the Bing spell-checker. Then we used the unigram matching, bigram matching, and 1-skip bigram matching modules for monolingual FAQ retrieval. For cross-lingual system, we used the following three modules: an SMS-to-English query translation system, an English-to-Hindi translation system, and cross-lingual FAQ retrieval.

A review of artificial intelligence and biologically inspired computational approaches to solving issues in narrative financial disclosure

Indisputably, financial reporting has a key role to play in the efficient workings of capitalist economies. Problems related to agency and asymmetric information (Jensen and Meckling, 1976) would abound and cripple financial markets, as it has done when left unchecked (Enron, WorldCom and Tyco). However for too long, quantitative data has monopolised the assessment and prediction role within this arena and this has contributed to the failures, borne out by research (Kumar & Ravi, 2007). As qualitative data proliferates, containing value relevant information it needs to be factored into the analysis. This paper reviews work on financial narrative disclosures and looks at conventional artificial intelligence and more recent biologically inspired computational approaches to catapult the domain to more progressive methods of using linguistic data in evaluations.

EmoSenticSpace: Dense Concept-Based Affective Features with Common-Sense Knowledge

This chapter proposes EmoSenticSpace, a new framework for affective common-sense reasoning that extends WordNet-Affect and SenticNet by providing both emotion labels and polarity scores for a large set of natural language concepts. The framework is built by means of fuzzy c-means clustering and support-vector-machine classification, and takes into account a number of similarity measures, including point-wise mutual information and emotional affinity. EmoSenticSpace was tested on three emotion-related natural language processing tasks, namely sentiment analysis, emotion recognition, and personality detection. In all cases, the proposed framework outperforms the state-of-the-art. In particular, the direct evaluation of EmoSenticSpace against psychological features provided in the benchmark ISEAR dataset shows a 92.15