Satire is prominent in user-generated content on various online platforms in the form of satirical news, customer reviews, blogs, articles, and short messages that are typically of an informal nature. As satire is also used to disseminate false information on the Internet, its computational detection has become a well-known issue. Existing work focuses primarily on formal document- or sentence-level textual data, whereas informal short texts have gotten less attention for satire detection. This paper presents a new model called BiLSTM self-attention (BiSAT) for detecting satire in informal short texts. It consists of various components such as input, embedding, self-attention, and two bi-directional long short-term memory (BiLSTM) layers for learning crucial contextual information pertaining to the satire present in the texts. The input layer uses the text as input to create an input vector, which is then given to the embedding layer to create the appropriate numeric vector. The output of the embedding layer is passed on to the first BiLSTM layer, which extracts contextual information-based sequences in the opposite direction. Between the first and second BiLSTM layers, a self-attention layer is employed to draw attention to the important satirical information that is acquired by the hidden layer of the first BiLSTM. The BiSAT model also takes a classic feature engineering approach, employing a 13-dimensional auxiliary feature vector comprised of features from four separate feature categories: sentiment, punctuation, hyperbole, and affective. The proposed BiSAT model is empirically evaluated on two benchmark datasets and a newly created dataset called Satire-280. It outperforms existing research and baseline methods by a significant margin. The Satire-280 dataset along with code can be downloaded from GitHub repository: https://github.com/Ashraf-Kamal/Satire-Detection.

Social networks are rapidly expanding due to their imperative role in disseminating information in a split second, emerging as the primary source for breaking news. As a result, the rich, user-generated information entices researchers to delve deeper and extract valuable insights. Event detection in online social networks (OSNs) is a research problem that has shifted researchers attention from traditional news media to online social media data. Event detection in OSNs is an automated process, addressing the impractical task of manually filtering potential events from vast amounts of online data. Unfortunately, the informality and semantic sparsity of online social networking text pose significant challenges to the event detection task. To this end, we present an approach namedconteNXtfor detecting events from Twitter (currently “X”) posts (also known as Tweets). To handle large amounts of data, the proposed method divides tweets into bins and uses postprocessing methods to extract bursty keyphrases. These keyphrases are then used to generate a weighted keyphrase graph using the Word2Vec model. Finally, Markov clustering is employed to cluster and detect events in the bursty keyphrase graph.conteNXtis evaluated on theEvent-Corpus2012benchmark dataset and two additional datasets extracted from the archive,Archive2020andArchive2021, using performance evaluation metrics: #events, precision, recall, and F1-score. The proposed approach outperforms state-of-the-art methods, includingSEDTWik,Twevent,Sentence-BERT,MABED,EDED,CommunityINDICATOR, andEventX. Additionally, the proposed approach is capable of detecting vital events that are not identified by the aforementioned state-of-the-art methods.

In this paper, we present a multi-task learning framework consisting of two interrelated components for the joint modeling of stance classification and rumor veracity prediction on Twitter. The proposed hierarchical framework models a conversation sequence using a graph attention network, leveraging a BERT-based word representation augmented with user credibility information. The lower component of the framework models the conversation sequence of a claim in addition to the BERT-based user content representation to predict the stance of the underlying tweets. It employs a modified graph attention network, which models a conversation thread by finding tweets path-to-root conversation sequences. Further, the learned stance representation is augmented with the users' credibility information and content representation to predict rumor veracity. The experimental evaluation results over two benchmark datasets show that the proposed approach outperforms the state-of-the-art methods.

In this paper, we discuss the issue of features that emerge during the prediction phase of a machine learning model, termed as unseen features. Because unseen features are absent from the vocabulary of the trained model, they are often rejected during the preprocessing stage of the learning model in standard machine learning approaches. We introduce the idea of unseen features and a method for identifying and using them for classification tasks. Because the dimension of feature vector required for trained machine learning model is going to differ upon incorporating unseen features of the testing data sample, it is not practical to directly incorporate unseen features since they only exist during the prediction phase of a machine learning model. As a result, the feature space for the training set is transformed to the embedding space which facilitates the use of unseen features. The proposed approach is empirically evaluated using standard metrics over three benchmark datasets in diverse circumstances (natural and balanced datasets) and on various text types – long-texts (aka structured texts) and short-texts (aka unstructured texts) considering five distinct classification algorithms. The experimental findings confirm the effectiveness of using unseen features during a machine learning model's deployment phase. The proposed unseen features enhanced technique outperforms the conventional approaches in both balanced class distribution and natural class distribution scenarios by a significant margin of at least 10%.

Domain-specific keyword extraction is a vital task in the field of text mining. There are various research tasks, such as spam e-mail classification, abusive language detection, sentiment analysis, and emotion mining, where a set of domain-specific keywords (aka lexicon) is highly effective. Existing works for keyword extraction list all keywords rather than domain-specific keywords from a document corpus. Moreover, most of the existing approaches perform well on formal document corpuses but fail on noisy and informal user-generated content in online social media. In this article, we present a hybrid approach by jointly modeling the local and global contextual semantics of words, utilizing the strength of distributional word representation and contrasting-domain corpus for domain-specific keyword extraction. Starting with a seed set of a few domain-specific keywords, we model the text corpus as a weighted word-graph. In this graph, the initial weight of a node (word) represents its semantic association with the target domain calculated as a linear combination of three semantic association metrics, and the weight of an edge connecting a pair of nodes represents the co-occurrence count of the respective words. Thereafter, a modified PageRank method is applied to the word-graph to identify the most relevant words for expanding the initial set of domain-specific keywords. We evaluate our method over both formal and informal text corpuses (comprising six datasets), and show that it performs significantly better in comparison to state-of-the-art methods. Furthermore, we generalize our approach to handle the language-agnostic case, and show that it outperforms existing language-agnostic approaches.

Mammalian Target of Rapamycin (mTOR) is a Ser/Thr protein kinase, and its role is integral to the autophagy pathway in cancer. Targeting mTOR for therapeutic interventions in cancer through autophagy pathway is challenging due to the dual roles of autophagy in tumor progression. The architecture of mTOR reveals two complexes - mTORC1 and mTORC2, each having multiple protein subunits. mTOR kinase inhibitors target the structurally and functionally similar catalytic subunits of both mTORC1 and mTORC2. In this paper, we have explored two different categories of molecular features - descriptors and fingerprints for developing predictive models using machine learning techniques. Random Forest variable importance measures and autoencoders are used to identify molecular descriptors and fingerprints, respectively. We have built various predictive models using identified features and their combination for predicting mTOR kinase inhibitors. Finally, the best model based on the Mathew correlation co-efficient value over the validation dataset is selected for screening kinase SARfari bioactivity dataset. In this study, we have identified twenty best performing descriptors for predicting mTOR kinase inhibitors. To the best of our knowledge, it is the first study on integrating traditional machine learning and deep learning-based approaches for feature extraction to predict mTOR kinase inhibitors.

Online Social Networks (OSNs) are witnessing sophisticated cyber threats, that are generally conducted using fake or compromised profiles. Automated agents (aka socialbots), a category of sophisticated and modern threat entities, are the native of the social media platforms and responsible for various modern weaponized information-related attacks, such as astroturfing, misinformation diffusion, and spamming. Detecting socialbots is a challenging and vital task due to their deceiving character of imitating human behavior. To this end, this paper presents an attention-aware deep neural network model, DeepSBD, for detecting socialbots on OSNs. The DeepSBD models users' behavior using profile, temporal, activity, and content information. It jointly models OSN users' behavior using Bidirectional Long Short Term Memory (BiLSTM) and Convolutional Neural Network (CNN) architectures. It models profile, temporal, and activity information as sequences, which are fed to a two-layers stacked BiLSTM, whereas content information is fed to a deep CNN. We have evaluated DeepSBD over five real-world benchmark datasets and found that it performs significantly better in comparison to the state-of-the-arts and baseline methods. We have also analyzed the efficacy of DeepSBD at different ratios of socialbots and benign users and found that an imbalanced dataset moderately affects the classification accuracy. Finally, we have analyzed the discrimination power of different behavioral components, and it is found that both profile characteristics and content behavior are most impactful, whereas diurnal temporal behavior is the least effective for detecting socialbots on OSNs.

The frequent usage of figurative language on online social networks, especially on Twitter, has the potential to mislead traditional sentiment analysis and recommender systems. Due to the extensive use of slangs, bashes, flames, and non-literal texts, tweets are a great source of figurative language, such as sarcasm, irony, metaphor, simile, hyperbole, humor, and satire. Starting with a brief introduction of figurative language and its various categories, this article presents an in-depth survey of the state-of-the-art techniques for computational detection of seven different figurative language categories, mainly on Twitter. For each figurative language category, we present details about the characterizing features, datasets, and state-of-the-art computational detection approaches. Finally, we discuss open challenges and future directions of research for each figurative language category.

In this paper, we present a graph-theoretic embedding-based approach to model user-generated contents in online social media for rumor detection. Starting with a small set of seed rumor words of four different lexical categories, we generate a words co-occurrence graph and apply centrality-based analysis to identify prominent rumor characterizing words. Thereafter, word embedding is applied to represent each category of seed words as numeric vectors and to train three different classification models for rumor detection. The performance of the proposed approach is empirically evaluated over two versions of a benchmark dataset. The proposed approach is also compared with one of the state-of-the-art methods for rumor detection and performs significantly better.

In this paper, we present a language-neutral graph-based sentiment analysis approach, SentiLangN, which uses character n-gram graph for modelling textual data to handle language-neutral unstructured expressions and noisy data. Since ordering and positioning of characters and words in a document plays a vital role in content analysis, the SentiLangN employs the longest common subsequence and degree similarity to capture inherent semantics of the textual data.

Due to increasing popularity of Bitcoin and other cryptocurrencies, proliferation of deceptive cryptocurrencies over the internet is a global concern. In this paper, we have identified a set of 24 features through analyzing Cryptocurrency Market Capitalization (CMC) data and propose a Multilayer Perceptron (MLP) architecture for detecting deceptive cryptocurrencies. The proposed MLP architecture is compared with three traditional machine learning algorithms over a real cryptocurrency dataset crawled from CMC website, and it performs significantly better.

Twitter, a microblogging online social network, is one of the most popular information sharing and communication platform. The large user-base and users mutual interactions generate massive amount of data that are rich source of information for predictive modeling, sentiment analysis, opinion mining, and other text information processing tasks. Understanding context embedded within text corpus and generating a contextual summary of the corpus is one of the promising research directions in the field of data analytics. In this paper, we present a layered graph-based approach using both content and structural data to analyze and summarize tweets at different levels of granularity. The proposed approach models tweets as a multi-dimensional graph and applies random walk to identify most informative tweets, which are further processed using a graph-theoretic approach, LexRank, to identify most informative sentences for summary generation. Finally, the summary texts are analyzed using TextRank algorithm to identify prominent keywords conceptualizing the context of the underlying corpus. The proposed summary generation and context learning approach is evaluated over four different real-world Twitter datasets using standard information retrieval metrics.

Sarcasm is a special category of figurative language, which is mainly used in online social media to convey messages with implicit semantics and criticism. Such messages are used for sarcastic remarks using contemptuous, ridicule, bitter, taunt, and mock related words or phrases. Though sarcasm detection is a well-considered problem by the researchers, to the best of our knowledge, none of them has considered the problem of self-deprecating sarcasm, which is a special category of sarcasm, mainly used by the users to deprecate or criticize themselves using sarcastic phrases. In this paper, we propose a novel self-deprecating sarcasm detection approach using an amalgamation of rule-based and machine learning techniques. The rule-based techniques aim to identify candidate self-around tweets, whereas machine learning techniques are used for feature extraction and classification. A total number of 11 features, including six self-deprecating features and five hyperbolic features are identified to train three different classifiers - decision tree, naïve Bayes, and bagging. The proposed approach is evaluated over a Twitter dataset containing 107536 tweets, and compared with some state-of-the-art methods for sarcasm detection.

Twitter is one of the most popular microblogging services, which is generally used to share news and updates through short messages restricted to 280 characters. However, its open nature and large user base are frequently exploited by automated spammers, content polluters, and other ill-intended users to commit various cybercrimes, such as cyberbullying, trolling, rumor dissemination, and stalking. Accordingly, a number of approaches have been proposed by researchers to address these problems. However, most of these approaches are based on user characterization and completely disregarding mutual interactions. In this paper, we present a hybrid approach for detecting automated spammers by amalgamating community-based features with other feature categories, namely metadata-, content-, and interaction-based features. The novelty of the proposed approach lies in the characterization of users based on their interactions with their followers given that a user can evade features that are related to his/her own activities, but evading those based on the followers is difficult. Nineteen different features, including six newly defined features and two redefined features, are identified for learning three classifiers, namely, random forest, decision tree, and Bayesian network, on a real dataset that comprises benign users and spammers. The discrimination power of different feature categories is also analyzed, and interaction- and community-based features are determined to be the most effective for spam detection, whereas metadata-based features are proven to be the least effective.

With the advent of huge data management systems storing voluminous data, there arises a need to develop efficient data analytics techniques for knowledge discovery at different levels of granularity. Resource Description Framework (RDF), mainly developed for Semantic Web, is presumably a good option when considering graph databases dealing with huge real-world data. RDF models information in the form of triples , and is considered as a useful tool to store graph data (aka linked data) where each edge can be stored as a triple. Due to existence of huge amount of linked data, mostly in the form of graphs, graph mining has been successful in attracting researchers from different research fields for efficient handling (storage, indexing, retrieval, etc.) of graph data. As a result, various APIs like GraphX and GraphFrames are developed to facilitate relational queries over graph data. Though GraphX is older than GraphFrames and processing SPARQL queries over GraphX has been explored by some researchers, to the best of our knowledge, SPARQL query processing over GraphFrames has not been explored yet. In this paper, we present an initial study on query-specific search space pruning and query optimization approach to process SPARQL queries over GraphFrames in an efficient manner. The experimental results, in terms of low response time for query execution, are encouraging, and give way to invest more research efforts in this direction.

Online social networks are facing serious threats due to presence of human-behaviour imitating malicious bots (aka socialbots) that are successful mainly due to existence of their duped followers. In this paper, we propose an approach to categorize Twitter users into three groups - active, reactive, and inactive targets, based on their interaction behaviour with socialbots. Active users are those who themselves follow socialbots without being followed by them, reactive users respond to the following socialbots by following them back, whereas inactive users do not show any interest against the following requests from anonymous socialbots. The proposed approach is modelled as both binary and ternary classification problem, wherein users' profile is generated using static and dynamic components representing their identical and behavioural aspects. Three different classification techniques viz Naive Bayes, Reduced Error Pruned Decision Tree, and Random Forest are used over a dataset of 749 users collected through live experiment, and a thorough analyses of the identified users categories is presented, wherein it is found that active and reactive users keep on frequently updating their tweets containing advertising related contents. Finally, feature ranking algorithms are used to rank identified features to analyse their discriminative power, and it is found that following rate and follower rate are the most dominating features.

Community detection is an important task for identifying the structure and function of complex networks. The task is challenging as communities often show overlapping and hierarchical behavior, i.e., a node can belong to multiple communities, and multiple smaller communities can be embedded within a larger community. Moreover, real-world networks often contain communities of arbitrary size and shape, along with outliers. This paper presents a novel density-based overlapping community detection method, OCMiner, to identify overlapping community structures in social networks. Unlike other density-based community detection methods, OCMiner does not require the neighborhood threshold parameter (ε) to be set by the users. Determining an optimal value for ε is a longstanding and challenging task for density-based clustering methods. Instead, OCMiner automatically determines the neighborhood threshold parameter for each node locally from the underlying network. It also uses a novel distance function which utilizes the weights of the edges in weighted networks, besides being able to find communities in un-weighted networks. The efficacy of the proposed method has been established through experiments on various real-world and synthetic networks. In comparison to the existing state-of-the-art community detection methods, OCMiner is computationally faster, scalable to large-scale networks, and able to find significant community structures in social networks.

The growing popularity of online social media is leading to its widespread use among the online community for various purposes. In the recent past, it has been found that the web is also being used as a tool by radical or extremist groups and users to practice several kinds of mischievous acts with concealed agendas and promote ideologies in a sophisticated manner. Some of the web forums are predominantly being used for open discussions on critical issues influenced by radical thoughts. The influential users dominate and influence the newly joined innocent users through their radical thoughts. This paper presents an application of collocation theory to identify radically influential users in web forums. The radicalness of a user is captured by a measure based on the degree of match of the commented posts with a threat list. Eleven different collocation metrics are formulated to identify the association among users, and they are finally embedded in a customized PageRank algorithm to generate a ranked list of radically influential users. The experiments are conducted on a standard data set provided for a challenge at ISI-KDD’12 workshop to find radical and infectious threads, members, postings, ideas, and ideologies. Experimental results show that our proposed method outperforms the existing UserRank algorithm. We also found that the collocation theory is more effective to deal with such ranking problem than the textual and temporal similarity-based measures studied earlier.

In this paper, we propose a unified framework, HOCTracker, for tracking the evolution of hierarchical and overlapping communities in online social networks. Unlike most of the dynamic community detection methods, HOCTracker adapts a preliminary community structure towards dynamic changes in social networks using a novel density-based approach for detecting overlapping community structures, and automatically tracks evolutionary events like birth, growth, contraction, merge, split, and death of communities. It uses a novel and efficient log-based approach to map evolutionary relations between communities identified at two consecutive time-steps of a dynamic network, which considerably reduces the number of community comparisons. Moreover, it does not require an ageing function to remove old interactions for identifying community evolutionary events. HOCTracker is applicable to directed/undirected and weighted/unweighted networks. Experimental results have shown that community structures identified by HOCTracker on some well-known benchmark networks are significant and in general better that the community structures identified by the state-of-the-art methods.

The popularity of Online Social Networks (OSNs) is often faced with challenges of dealing with undesirable users and their malicious activities in the social networks. The most common form of malicious activity over OSNs is spamming wherein a bot (fake user) disseminates content, malware/viruses, etc. to the legitimate users of the social networks. The common motives behind such activity include phishing, scams, viral marketing and so on which the recipients do not indent to receive. It is thus a highly desirable task to devise techniques and methods for identifying spammers (spamming accounts) in OSNs. With an aim of exploiting social network characteristics of community formation by legitimate users, this paper presents a community-based framework to identify spammers in OSNs. The framework uses community-based features of OSN users to learn classification models for identification of spamming accounts. The preliminary experiments on a real-world dataset with simulated spammers reveal that proposed approach is promising and that using community-based node features of OSN users can improve the performance of classifying spammers and legitimate users.

This paper aims to identify user-centric features to calculate the strength of social ties between Online Social Network (OSN) users, and models the same using Latent Space Model (LSM). The modeling approach processes a socio-centric user-set as the users are directly (friend) or indirectly (friend-of-friend) related to a seed (target) user, which makes it easier to identify social ties between users as compared to random sampling from a set of diverse OSN users. For a given user, interaction data up to two levels is modeled and analyzed to generate a user-centric social network. Eleven different features related to Facebook have been identified to calculate the strength of social ties between users. LSM is used to visualize relationships in user-centric historical data and to estimate the probability of social ties between OSN users. The users are plotted using LSM in a three-dimensional (3D) social space around a seed user, and a link probability function is devised to calculate the probability of link between any two users with respect to the persona of the seed user. A sphere of influence around each user demarcating its active influence area is also identified and discussed in this paper.

In this paper, we propose a unified framework OCTracker for tracking overlapping community evolution in online social networks. OCTracker adapts a preliminary community structure towards dynamic changes in social networks using a novel density-based approach for detecting overlapping community structures and automatically detects evolutionary events like birth, growth, contraction, merge, split, and death of communities with time. Unlike other density-based community detection methods, the proposed method does not require the neighborhood threshold parameter to be set by the users, rather it automatically determines the same for each node locally.

In this paper, we present a novel agglomerative clustering method to identify cliques in dark Web forums. Considering each post as an individual entity accompanying all the information about its thread, author, time-stamp, etc., we have defined a similarity function to identify similarity between each pair of posts as a blend of their contextual and temporal coherence. The similarity function is employed in the proposed clustering algorithm to group threads into different clusters that are finally presented as individual cliques. The identified cliques are characterized using the homogeneity of posts therein, which also establishes the homogeneity of their authors and threads as well.

Short Message Service (SMS) has been increasingly exploited through spam propagation schemes in recent years. This paper presents a new method for graph-based learning and classification of spam SMS on mobile devices and smart phones. Our approach is based on modeling the content and patterns of SMS syntax into a direct ed-weighted graph through exploiting modern composition style of messages. The graph attributes are then used to classify spam messages in real-time by using KL-Divergence measure. Experimental results on two real-world datasets show that our proposed method achieves high detection accuracy with less false alarm rate to detect spam messages. Moreover, our approach requires relatively less memory and processing power, making it suitable to deploy on resource-constrained mobile devices and smart phones.

Over the past few years, Online Social Networks (OSNs) have emerged as cheap and popular communication and information sharing media. Huge amount of information is being shared through popular OSN sites. This aspect of sharing information to a large number of individuals with ease has attracted social spammers to exploit the network of trust for spreading spam messages to promote personal blogs, advertisements, phishing, scam and so on. In this paper, we present a Markov Clustering (MCL) based approach for the detection of spam profiles on OSNs. Our study is based on a real dataset of Facebook profiles, which includes both benign and spam profiles. We model social network using a weighted graph in which profiles are represented as nodes and their interactions as edges. The weight of an edge, connecting a pair of user profiles, is calculated as a function of their real social interactions in terms of active friends, page likes and shared URLs within the network. MCL is applied on the weighted graph to generate different clusters containing different categories of profiles. Majority voting is applied to handle the cases in which a cluster contains both spam and normal profiles. Our experimental results show that majority voting not only reduces the number of clusters to a minimum, but also increases the performance values in terms of FPand FBmeasures from FP=0.85 and FB=0.75 to FP=0.88 and FB=0.79, respectively.

In this paper, we propose a Markov Clustering (MCL) based text mining approach for namesake disambiguation on the Web. The novelty of the proposed technique lies in modeling the collection of web pages using a weighted graph structure and applying MCL to crystalize it into different clusters, each one containing the web pages related to a particular namesake individual. The proposed method focuses on three broad and realistic aspects to cluster web pages retrieved through search engines - content overlapping, structure overlapping, and local context overlapping. The efficacy of the proposed method is demonstrated through experimental evaluations on standard datasets.

Tag cloud, also known as word cloud, are very useful for quickly perceiving the most prominent terms embedded within a text collection to determine their relative prominence. The effectiveness of tag clouds to conceptualize a text corpus is directly proportional to the quality of the keyphrases extracted from the corpus. Although, authors provide a list of about five to ten keywords in scientific publications that are used to map them into their respective domain, due to exponential growth in non-scientific documents on the World Wide Web, an automatic mechanism is sought to identify keyphrases embedded within them for tag cloud generation. In this paper, we propose a web content mining technique to extract keyphrases from web documents for tag cloud generation. Instead of using partial or full parsing, the proposed method applies n-gram technique followed by various heuristics-based refinements to identify a set of lexical and semantic features from text documents. We propose a rich set of domain-independent features to model candidate keyphrases very effectively for establishing their keyphraseness using classification models. We also propose a font-determination function to determine the relative font-size of keyphrases for tag cloud generation. The efficacy of the proposed method is established through experimentation. The proposed method outperforms the popular keyphrase extraction system KEA.

In this paper, we present the design of a web content mining system to identify and extract aliases of a given entity from the Web in an automatic way. Starting with a pattern-based information extraction process, the system applies n-gram technique to extract candidate aliases. Thereafter, various statistical measures are applied to identify feasible aliases from them. The extracted aliases can be used to generate profiles of suspects and keep track of their movements on the Web using different identities.

Vulnerability exploits present in malformed messages are one of the major sources to remotely launch malicious activities in different protocols. Sometimes, a single malformed message could be enough to crash remote servers or to gain unfettered access over them. In this paper, we propose the design of a generic vulnerability exploits detection system xMiner to detect malformed messages in real time for avoiding any network hazard. The proposed xMiner exploits the information embedded within byte-level sequences of network messages. xMiner applies multi-order Markov process and principal component analysis (PCA) to extract novel discriminative features and uses them to detect attacks launched through malicious packets in real-time. The novelty of xMiner lies in its light-weight design which requires less processing and memory resources and makes it easily deployable on resource-constrained devices like smart phones. The system is evaluated on real-world datasets pertaining to three different protocols - HTTP, FTP and SIP. Five different classifiers are deployed to establish the effectiveness of the proposed system. On evaluation we found that the decision tree classifier performs well for HTTP and FTP datasets whereas, SVM shows highest performance in case of SIP packets.

Ontologies represent a method of formally expressing a shared understanding of information, and have paved the way for sharing concepts across applications in an unambiguous way. However, these ontologies are assumed to be hand-crafted, pre-defined structures with crisp concept descriptions and inter-concept relations. Crisp definitions however are not sufficient for real-world applications like ontology-based information extraction from unstructured text. In this paper we propose an enhancement of the ontology structure to a fuzzy ontology, which provides a mechanism to store imprecise concept definitions. The proposed fuzzy ontology framework can also help in ascertaining similarities and dissimilarities of concept definitions across distributed ontologies representing the same domain. Our design of fuzzy ontology is motivated by fuzzy set theoretic representation and reasoning, and is entirely different from other fuzzy ontology designs, which use only co-occurrence of concepts to determine their closeness. We have cited examples from various domains to show the necessity and capability of the structure in representing real-world knowledge.

Ontologies are proposed as a means for knowledge sharing among applications but, it is often not possible to converge to a single unambiguous ontology that is acceptable to all knowledge engineers. Different ontologies vary greatly in terms of the level of detail of their representations, as well as the nature of their underlying logical specifications. Interoperability among different ontologies becomes essential to gain from the power of the existing domain ontologies. In this paper we have proposed a fuzzy ontology framework in which a concept descriptor is represented as a fuzzy relation which encodes the degree of a property value using a fuzzy membership function. Other than concept descriptors, the semantic relations in the ontology like IS-A, HAS-PART etc. are also associated a strength of association. The strength of association between two concepts determines the "uniformity" with which these two concepts have been defined identically across different ontologies. The fuzzy ontology framework provides appropriate support for application integration by identifying the most likely location of a particular term in the ontology

Domain ontology can help in information retrieval from documents. But ontology is a pre-defined structure with crisp concept descriptions and inter-concept relations. However, due to the dynamic nature of the document repository, ontology should be upgradeable with information extracted through text mining of documents in the domain. This also necessitates that concepts, their descriptions and inter-concept relations should be associated with a degree of fuzziness that will indicate the support for the extracted knowledge according to the currently available resources. Supports may be revised with more knowledge coming in future. This approach preserves the basic structured knowledge format for storing domain knowledge, but at the same time allows for update of information. In this paper, we have proposed a mechanism which initiates text mining with a set of ontological concepts, and thereafter extracts fuzzy relations through text mining. Membership values of relations are functions of frequency of co-occurrence of concepts and relations. We have worked on the GENIA corpus and shown how fuzzy relations can be further used for guided information extraction from MEDLINE documents.