Ishan Verma

Extraction and Compilation of Events and Sub-events from Twitter

Twitter has emerged as a great source to provide insights about upcoming planned and unplanned events of social, economic and political relevance. Big events are publicized and known in advance, but smaller, unplanned sub-events around them are not always advertised. These unplanned events may have a large localized impact. If known in advance, knowledge about events like threats, protests, demonstrations etc. or even about large flash mobs can be utilized by planners and event managers. Given the large volumes of tweets floating around at any given time, identifying relevant sub-events is a non-trivial task. In this paper, we explore machine learning techniques to identify, extract and build a map of small sub-events around a big, popular event. We use CRFs to extract event components from tweets. Events are resolved for uniqueness and compiled into a complete calendar. The model is evaluated on tweets around Olympic Games. The framework is generic enough to be adapted to other domains.

Event Detection from Business News

An event is usually defined as a specific happening associated with a particular location and time. Though there has been a lot of focus on detecting events from political and other general News articles, there has not been much work on detecting Business-critical events from Business News. The major difference of business events from other events is that business events are often announcements that may refer to future happenings rather than happenings that have already occurred. In this paper, we propose a method to identify business-critical events within News text and classify them into pre-defined categories using a k-NN method. We also present an event-based retrieval mechanism for business News collections.

Contextually Intelligent Recommendation of Documents from User-Subscribed Channels

There are several professional, academic and business channels catering to the information needs of professionals. In this paper we have presented a contextual recommendation system that gathers documents from these channels on behalf of a user and recommends the most relevant documents from the collection based on the current work-context of the user. Starting with an initial context, the system employs reinforcement learning to understand user interests and then recommend contextually relevant articles only and thereby reduce information overload. The paper presents a theoretical framework for context detection and contextual recommendation and some experimental results from simulated environments.

Text-Driven Multi-structured Data Analytics for Enterprise Intelligence

Text data constitutes the bulk of all enterprise data. Text repositories are not only tacit store-houses of knowledge about its people, projects and processes but also contain invaluable information about its customers, competitors, suppliers, partners and all other stakeholders. Mining this data can provide interesting and valuable insights provided it is appropriately integrated with other enterprise data. In this paper we propose a framework for text-driven analysis of multi-structured data.

DemandMiner: Extraction and Visualization of Product Demands from Open Web

In this paper, we present the DemandMiner, an online tool that leverages Natural Language Processing and Machine Learning techniques to extract demand related to different industries from large volume of News articles. We also propose techniques to enrich the information components extracted from a document by associating them with their sense oriented entities. The enriched components are then analyzed for generating informed insights to generate contextual reports that can convey greater sense to the decision-makers, analysts and knowledge workers than simple display of events. An interactive visualization system is also provided for searching and querying the underlying collection of data and the derived insights.

Multi-structured Data Analytics using Interactive Visualization to aid Business Decision Making

Intelligent decision making highly relies on the ability to combine data from multiple sources from both within and outside the organization and generate appropriate insights. Multi-structured data analytics holds special significance for organizations, where different types of data are produced by different sources and applications, ad assimilating information from this diverse collection poses huge challenges. Data collection can contain information in the form of numbers, aggregates, time-series, reports, proposals, logs, communication records etc. Insight generation in such a scenario requires novel methodologies to judiciously link information from multiple sources and reason with them. This paper focuses on developing methodologies for intelligent integration of multi-structured data. The emphasis is to generate multiple superimposed data visualizations using intelligent methods of linking information amongst data from unstructured and structured sources and also to facilitate interactive data explorations through drill down facilities for interesting pieces of information.

A Semantic Overlapping Clustering Algorithm for Analyzing Short-Texts

The rise in volumes of digitized short-texts like tweets or customer complaints and opinions about products and services pose new challenges to the established methods of text analytics both due to the sparseness of text and noise. In this paper we present a new semantic clustering algorithm, which first discovers frequently occurring semantic concepts within a repository, and then clusters the documents around these concepts based on concept distribution within them. The method produces overlapping clusters which generates far more accurate view of content embedded within real-life communication texts. We have compared the clustering results with LSH based clustering and show that the proposed method produces fewer overall clusters with more semantic coherence within a cluster.

Fuzzy Multi-label Classification of Customer Complaint Logs Under Noisy Environment

Analyzing and understanding customer complaints has become and important issue in almost all enterprises. With respect to this, one of the key factors involve is to automatically identify and analyze the different causes of the complaints. A single complaint may belong to multiple complaint domains with fuzzy associations to each of the different domains. Thus, single label or multi-class classification techniques may not be suitable for classification of such complaint logs. In this paper, we have analyzed and classified customer complaints of some of the leading telecom service providers in India. Accordingly, we have adopted a fuzzy multi-label text classification approach along with different language independent statistical features to address the above mentioned issue. Our evaluation shows combining the features of point-wise mutual information and unigram returns the best possible result.