The aim of this project is to develop a multi-model designed computational model to forecast pain change for individual patients and provide optimal treatment recommendations.
The aim of this study is to detect changes in behavior (agitation, depression, and apathy) and activity patterns of patients with dementia and their caregivers.
Knowledge Graph Development Platform
Dynamic Knowledge Graphs and Adaptive Knowledge Networks
Change is a law of nature, and static KGs like DBpedia fail to capture this dynamic flow of information. Diverse applications of AI are increasingly relying on the knowledge which is not necessarily static e.g., President_of_the_USA, champion_of_FIFA_World_Cup are temporally-sensitive facts, unlike birth_date or death_date. The need for accurate temporal query responses by dominant search engines requires extracting, maintaining, and updating the temporal facts in KGs. Analyzing real-world dynamic events (e.g., elections, natural disasters, etc.) requires real-time predictive analysis, trend analysis, spatiotemporal decision making, and public opinion analysis. In this project, we propose to curate Adaptive Knowledge Networks from incoming real-time multimodal spatiotemporally evolving data which change with time.
Scalable Interface for Biomedical Knowledge Graph Creation (ScIKG)
With the dramatically growing volume of information in the biomedical domain, there is an unmet need to extract contextually- and semantically- important information. Existing approaches focus on entity extraction to enrich the query but do not incorporate context. We extend entity recognition task for the biomedical domain by leveraging contextually related n-grams (CRng). The essence of CRng arises from the fact that the semantics of an entity is defined by the context in which it is used. We propose Scalable Interface for Biomedical Knowledge Graph (ScIKG), a platform for extraction, enrichment, and annotation of social media data using existing biomedical knowledge bases and biomedical literature. It provides users information that is contextually relevant to the queries. It provides medical domain experts an insight into the general public awareness in their domain. Simultaneously, it provides a window for a naive user to access multiple social media sources along with latest research works in the biomedical domain.
Social media enhanced Collective Intelligence
We propose a community detection and characterization algorithm that incorporates the contextual information of node attributes described by multi- ple domain-specific hierarchical concept graphs. The core problem is to find the context that can best summarize the nodes in com- munities, while also discovering communities aligned with the context summarizing communities. We formulate the two intertwined problems, optimal community-context computation, and community discovery, with a coordinate-ascent based algorithm that iteratively updates the nodes’ community label assignment with a community-context and computes the best context summa- rizing nodes of each community. Our unique contributions include (1) a composite metric on Informativeness and Purity criteria in searching for the best context summarizing nodes of a community; (2) a node similarity measure that incorporates the context-level sim- ilarity on multiple node attributes; and (3) an integrated algorithm that drives community structure discovery by appropriately weigh- ing edges.
Context-Aware Harassment Detection on Social Media
Development of harassment concepts ontology (schema) and extraction of instances: Harassment knowledge graph.
Understanding the temporal evolution of relationships between people is critical to identify the difference between sarcasm and harassment. Because harassment is a temporally sensitive phenomenon, Dynamic Evolving Knowledge Graphs (DEKG) provide a way to represent the evolution of time-sensitive entities and relationships in a machine-readable format. In order to provide a solution that can benefit this project and is potentially reusable, we have developed and used a generic platform for representing and querying knowledge graphs. We have harvested knowledge from DBpedia and YAGO (more generally, structured, semi-structured, and curated data sources). In order to capture, critical real-world temporal evolution of entities, relationships and events, we have explored context-based representation and reasoning issues relevant to (DEKG). In summary, we expect this foundational work to assist in explicit representation of various forms of harassment and their instances, reason about the changes and intensity over time, and permit analysis that sheds light on group-based behaviors.
KHealth: Semantic Multisensory Mobile Approach to Personalized Asthma Care
kHFilos: Knowledge driven mHealth-Know your health !!
The healthcare medical records of asthma patients hold information such as their symptoms, diagnosis and respective treatments. These clinical notes are cumbersome to comprehend and they cannot record patient’s health progression accurately based on periodic clinical visits. To bridge this gap, we developed kHealth kit to continuously collect Patient Generated Health Data (PGHD). kHFilos is a knowledge-enabled mHealth service for self management and self appraisal based on multimodal data collected from kHealth kit2 and data from clinical notes. It generates a personalized disease progression score, patient compliance score, symptom comparison based on clinical notes by utilizing knowledge from the domain expert, to understand the patient’s health with ease.
Kno.e.sis Alchemy for Heathcare
Unsupervised Biomedical Relation Extraction with Knowledge bases and word Embeddings
Extracting biomedical entities and relationships from text is an important task and a key to many biomedical applications including undiscovered public knowledge. Existing work on entity, relation extraction has primarily relied on supervised and semi-supervised learning techniques which require large amounts of human annotated data. The creation of training data can be difficult due to the different linguistic styles (formal versus informal) in the text and the complexity of the biomedical entities (simple versus compound entities). In this work, we propose an unsupervised extraction of biomedical entities and relationships by leveraging existing biomedical knowledge bases such as SNOMED, ICD-10 and PubMed. The proposed technique exploits probabilistic language model for identifying candidate compound entities and generates context-based embeddings for candidate entities using the biomedical knowledge bases. Finally, we use hierarchical clustering over the context-based embeddings for relation extraction. We evaluate our work on a public dataset BioInfer, a unique gold standard corpus providing annotation of biomedical entities and relationships. The corpus is focussed towards protein, gene and RNA relationships and entities, which makes it unique and complex.