Sailesh Kumar Sathish

Evolving the User Graph: From unsupervised topic models to knowledge assisted networks

The next generation intelligent devices need to understand and evolve with the user. Towards this goal, we present a User Graph generation framework that models users level of interest and knowledge across a set of categories. The user graph is built through an unsupervised and semi-supervised topic modeling process, using latent semantic analysis technology. The self-evolving framework utilizes in-device user data, is built and managed within a local mobile device, thereby ensuring user privacy without the need for additional network based infrastructure. We present and analyze our trial results, aimed at optimizing model accuracy and execution efficiency. In addition to native application adaptation use cases, we also present three new services: Graph Clusters, Graph Shares and Graph Nets that utilize the framework.

Med-Tree: A user knowledge graph framework for medical applications

The field of context and intelligence, as a topic of pervasive computing, has been gaining considerable momentum. Typically, context-aware intelligence is applied to understand the situation of the users and their behavior with the objective of providing adaptive services that are closely associated with that context. In this work, we have taken an orthogonal approach wherein we attempt to aggregate knowledge and cognition, of the user, on a given topic to build models out of them. The model thus created is analyzed to derive inferences about the user, where the analysis is performed on a graph model comprising topics based information obtained by mining domain specific personal data sources and from certain facts on which the user has expressed fair level of belief. We have explored the possibility of deriving beneficial information by provisioning an appropriate representation of knowledge as belief-graph with specific orientation in healthcare and call this model as Med-Tree. Subject to privacy conditions, we open up the belief-graph model to establish objective based social connections that gets contextually bound. As a next step, such contextually bound ad-hoc networks are subjected to advanced querying process resulting in useful information extraction and inferences. Leveraging on users knowledge or the belief-graph, the proposed Med-Tree could help derive benefits towards better personal healthcare and disease management.

Enabling Multi-Topic and Cross-Language Browsing Using Web-Semantics Service

Mobile web interaction using multiple modalities such as speech, gesture and touch is a well-researched area with examples of multiple framework deployments within the domain. Here, we explore usage of semantics of web content as a new modality for mobile web content discovery and navigation. In addition to semantics-based interactivity, the framework enables dynamic cross-language switching between web pages represented in different languages representing same topics through use of their inherent semantic relations. The approach uses transforming natural web content representations to an intermediate index of latent topic distributions. As the system is semantically indexed, no translation or transliteration techniques are required for cross language switch. The multi-dimensional topic index is built using content-topic distributions which in turn are inferred using supervised topic models. The end-to-end system has been developed as an industrial application with enhanced navigational capability provided as an extension to Samsung Browser v4.0 for Android platform. The service also allows standard browsers to query limited services using the web services RESTful API. The current system provides semantic navigation along with cross-language support for English, Hindi, and Spanish. The semantics based navigation modality, benchmarked against state of the art semantic browsers on objective and usability metrics by users demonstrate the superiority of this system. The cross-language system has been evaluated by language experts for both cross-mapping accuracy as well as relevance order mapping resulting in a subjective cumulative gain score of over 72%.

A Generic Framework for Recommendations Based on User Data Aggregation

The next generation intelligent devices need to grow and evolve with the user. Mobile user framework solution allows us to move towards this goal. In this paper, we describe a service framework that captures users interest and intent which is mined through latent analysis of content. Applications using such a framework can provide adaptive and customized services to the user. We provide details of four demonstrator applications that use this framework, utilizing user interest and intent information to provision targeted advertisements, recommend products and songs. We provide details of our evaluation with subjective and objective analysis of the system.

Semantics-Based Browsing Using Latent Topic Warped Indexes

Mobile web browsing has inherent usability and content discovery problems that warrants alternate browsing paradigms. To provision improved web usability, a server hosted multi-topic content indexing system and a client-based topic cluster solution has been developed that allows the user to browse the web based on content semantics alone. Semantic connections between content are established based on mono-topic and multi-topic indexes built by our service framework. Topic indexing is performed through latent analysis of web content and semantic relations between content is established through distance based relational mining. From user perspective, once an anchor page is loaded in mobile web browser, the user can browse semantically related pages with respect to the anchor page using intuitive vertical and horizontal swipe actions. The pages can be ordered temporally which adds a longitudinal (temporal) dimension to browsing. We present our experimental results based on subjective and objective evaluations via user trials validating semantics-based browsing paradigm.