M. P. S. Bhatia

Wireless sensor networks for monitoring the environmental activities

The area of sensor network has a long history and many kind of sensor devices are used in various real life applications. Here, we introduce Wireless sensor network which when combine with other areas then plays an important role in analyzing the data of forest temperature, bioinformatics, water contamination, traffic control, telecommunication etc. Due to the advancement in the area of wireless sensor network and their ability to generate large amount of spatial/temporal data, always attract researchers for applying data mining techniques and getting interesting results. Wireless sensor networks in monitoring the environmental activities grows and this attract greater interest and challenge for finding out the patterns from large amount of spatial/temporal datasets. These datasets are generated by sensor nodes which are deployed in some tropical regions or from some wearable sensor nodes which are attached with wild animals in wild life centuries. Sensor networks generate continuous stream of data over time. So, Data mining techniques always plays a vital role for extracting the knowledge form large wireless sensor network data. In this paper, we present the detection of sensor data irregularities, Sensor data clustering, Pattern matching and their interesting results and with these results we can analyze the sensor node data in different ways.

Data clustering with modified K-means algorithm

This paper presents a data clustering approach using modified K-Means algorithm based on the improvement of the sensitivity of initial center (seed point) of clusters. This algorithm partitions the whole space into different segments and calculates the frequency of data point in each segment. The segment which shows maximum frequency of data point will have the maximum probability to contain the centroid of cluster. The number of clusters centroid (k) will be provided by the user in the same manner like the traditional K-mean algorithm and the number of division will be k∗k (‘k’ vertically as well as ‘k’ horizontally). If the highest frequency of data point is same in different segments and the upper bound of segment crosses the threshold ‘k’ then merging of different segments become mandatory and then take the highest k segment for calculating the initial centroid (seed point) of clusters. In this paper we also define a threshold distance for each clusters centroid to compare the distance between data point and clusters centroid with this threshold distance through which we can minimize the computational effort during calculation of distance between data point and clusters centroid. It is shown that how the modified k-mean algorithm will decrease the complexity & the effort of numerical calculation, maintaining the easiness of implementing the k-mean algorithm. It assigns the data point to their appropriate class or cluster more effectively.

SVM classification to distinguish Parkinson disease patients

In this paper we have discussed the importance of data mining in the field of bioinformatics and various subfields of bioinformatics in which data mining has shown its great impact. Using a data mining tool, Weka, we pre- process the dataset on which we have worked and then using one of the classification methods i.e. Support Vector Machine method (SVM), we distinguished people with Parkinsons disease from the healthy people. Appling libsvm we have tried to find the best possible accuracy on different kernel values for the given dataset. We study the ROC curve variation, and the way the value of true positive and false positive rates changes with increasing number of the cross validation folds.

Tracking on-line radicalization using investigative data mining

The increasing complexity and emergence of Web 2.0 applications have paved way for threats arising out of the use of social networks by cyber extremists (Radical groups). Radicalization (also called cyber extremism and cyber hate propaganda) is a growing concern to the society and also of great pertinence to governments & law enforcement agencies all across the world. Further, the dynamism of these groups adds another level of complexity in the domain, as with time, one may witness a change in members of the group and hence has motivated many researchers towards this field. This proposal presents an investigative data mining approach for detecting the dynamic behavior of these radical groups in online social networks by textual analysis of the messages posted by the members of these groups along with the application of techniques used in social network analysis. Some of the preliminary results obtained through partial implementation of the approach are also discussed.

Exploiting grammatical dependencies for fine-grained opinion mining

In any sentence, words are arranged in a proper sequence to communicate information. The complete meaning of a sentence is not only determined by the meaning of words, but also by the pattern in which words are arranged. Essentially each word in a sentence possesses grammatical corporations with other words for correct utterance of meaning, such corporations between words is called binary grammatical relation or dependency (BGD). We believe that the analysis of a sentence on the basis of grammatical dependencies among the words is very useful for application such as opinion mining (OM) from a domain specific free format product reviews. Such free format reviews are largely available on Web due to presence of many e-commerce sites. Because of unavailability of universal resources (e.g. opinion word lexicon and feature corpus) for all application domains, OM becomes a very challenging job. OM from free format product reviews mainly deals with extraction of domain specific features, identification of opinion words corresponding to each features and determination of semantic orientation (SO) of opinion of words (e.g. large, small, loud, etc), which does not belong to set prior polarity opinion words (e.g. beautiful, excellent, good, etc.). In this paper, we discuss the role of BGDs among the words of a sentence for opinion mining and explore many BGDs that can directly facilitate the OM.

Statistical approach for community mining in social networks

The popularity of social networking on the Web and the explosive combination with data mining techniques open up vast and so far unexplored opportunities for social intelligence on the Web. A network community is a special sub-network that contains a group of nodes sharing similar linked patterns. Many community mining algorithms have been developed in the past. In this work, we have presented a new algorithm BFC (breadth first clustering) which uses statistical approach for community mining in social networks. The algorithm proceeds in breadth first way and incrementally extract communities from the network. This algorithm is simple, fast and can be scaled easily for large social networks. The effectiveness of this approach has been validated using network examples.

Contextual proximity based term-weighting for improved web information retrieval

Despite its success as a preferred or de-facto source of information, the Web implicates two key challenges: To provide improved systems that retrieve the most relevant information available, and, secondly, how to target search on information that satisfies users need with accurate balance of novelty and relevance. Nevertheless, Web content is not always easy to use. Due to the unstructured and semi-structured nature of the Web pages & design idiosyncrasy of Websites, it is a challenging task to organize & manage content from the Web. Web Mining tries to solve these issues that arise due to the WWW phenomenon. This paper proposes a novel context-based paradigm for improving Web Information Retrieval, given a multi-term query. The technique referred to as the Contextual Proximity Model (CPM), captures query context and matches it against term context in documents to determine term significance and topical relevance. It makes use of the co-information metric to detect the query context. This contextual evidence is used as an additional input to disambiguate and augment the users explicit query and dynamically contribute to the term frequency metric to ensure a vital, positive impact on retrieval accuracy.

Community Detection Approaches in Real World Networks: A Survey and Classification

Online social networks have been continuously evolving and one of their prominent features is the evolution of communities which can be characterized as a group of people who share a common relationship among themselves. Earlier studies on social network analysis focused on static network structures rather than dynamic processes, however, with the passage of time, the networks have also evolved and the researchers have started to focus on the aspect of studying dynamic behavior of networks. This paper aims to present an overview of community detection approaches graduating from static community detection methods towards the methods to identify dynamic communities in networks. The authors also present a classification of the existing dynamic community detection algorithms along the dimension of studying the evolution as either a two-step approach comprising of community detection via static methods and then applying temporal dynamics or a unified approach which comprises of dynamic detection of communities along with their evolutionary characteristics.

Quantifying modified opinion strength: A fuzzy inference system for Sentiment Analysis

In Sentiment Analysis or Opinion Mining, automatic quantification of the strength of opinion, expressed on any feature is very important task. However, it is quite challenging to automatically quantify the strength of opinion words, whenever they get modified by adverbial modifiers. For example the intensity of an opinion word “beautiful” gets increased in “very beautiful”, gets decreased in “slightly beautiful” and complemented in “not beautiful”. In our work, we have designed a fuzzy inference system based on experimentally designed fuzzy membership functions and concepts of hedges to standardized and formulate the process of strength quantification of subjective sentences when strength of opinion word get modified by the presence of n-gram adverbial modifiers pattern in the sentence. Using our membership functions and considering the manually quantified opinion strength for n-gram adverbial modifiers pattern as a baseline, we observed that our fuzzy inference system is producing output values with a very small average root mean square error with the targeted baseline points.

Analyzing Delhi Assembly Election 2015 Using Textual Content of Social Network

With the emergence of web 2.0, a large number of social networking sites (SNSs) have been evolved. Now-a-days, these social networking sites are attracting millions of users for sharing their views on various issues (e.g. politics, sports, products). We believe that, due to active participation of millions of users on social networking sites the SNSs forms a virtual world that has very close correspondence with the real world communities that possess immense possibilities to mirror the real world events and activities. Motivated with this, in this work we performed analysis of the textual content of Twitters data related to Delhi Assembly Election 2015 in a manner to predict election results. The main contributions of this work includes (i) Preparation and use of events and time specific training dataset to train the classifier for better accuracy (ii) Design of mapping functions that maps the Twitters sentiment share to the seat counts of top three contesting parties, with minimum root mean squared error (RSME) regardless of having lots of demographic diversities. The overall results are very close to ground reality, which strengthen our beliefs.