HealthAidKB:Extracting Procedural Knowledge Of Health Care From Web Communities

Empowering computers with machine-readable human knowledge has been a long-standing goal of artificial intelligence.This goal has seen big progress via advances in knowledge harvesting.In the past decades,automatic knowledge base construction has attracted much attention from the research community and resulted in significant progress in answering entity-centric queries.Knowledge graphs are at the core of the semantic search as they store facts about common entities.Today,publicly available knowledge graphs provide millions of entities(such as people,organizations,locations,and creative works like books,music,etc.)and billions of statements about the entities(such as who studied where,which country has which capital,or which singer performed which song).However,in situations where the user query is demanding procedural knowledge instead of entity related facts,the current knowledge graphs do not have enough resources to satisfy users' demand.On the other hand,much of the commonsense knowledge about the real world is in the form of procedures or sequences of actions.Studies of query analysis from the search log show that a significant amount of user queries involve task-oriented queries.Unfortunately,these kinds of knowledge are missing from most knowledge graphs and commonsense knowledge bases in use.To empower semantic search,and other intelligent applications,computers need to go beyond entities and encompass a much broader understanding of the world properties of everyday objects,human activities,and more.This work instead focuses on harvesting procedural knowledge to answer task-oriented queries.An example of such queries might be 'how to stop nausea using acupressure'or'how to aid digestion naturally'.To answer such questions,we proposed and implemented HelthAidKB:a procedural knowledge base of health-care.The main goal of this research work is to construct a machine-readable domain targeted high precision procedural knowledge base containing task frames from web communities and use it for downstream applications such as semantic search,search task suggestion,query expansion,and others.The main contribution of this thesis is as follows:(1)We constructed a procedural knowledge base of health-care,HealthAidKB.To construct HealthAidKB,we leveraged expert-curated semi-structured medical resources on the web.We developed a pipeline of methods leveraging open information extraction tools to extract procedural knowledge by tapping into online communities.First,we used web crawling techniques to extract domain targeted knowledge from on-line communities.To this end,we developed a python crawler program.The result of the crawler is given as input to the Open IE tool for extracting structured knowledge,which will give us the task frame.The extracted tasks are further processed in the pipeline by taxonomic induction and further attribute extraction for enhancing the structured knowledge.The knowledge organization stage in the pipeline has a clustering method that is used to canonicalize the task frames into clusters based on the similarity of the problems they intend to solve.To this end,we implemented hierarchical agglomerative clustering techniques.The resulting know-how knowledge base,HealthAidKB,consists of more than 71K task frames which are structured hierarchically and categorically;and can be used in many applications such as semantic search,digital personal assistants,human-computer dialogue,and computer vision.A comprehensive evaluation of our knowledge base shows high accuracy.(2)To showcase the usefulness of our knowledge base,we conducted two experiments;(i)answering task-oriented queries,and(ii)an extrinsic use-case of searching relevant YouTube videos for task-oriented queries.For the first problem,we investigated whether task-oriented queries can benefit from our procedural knowledge bases.To this end,we devised a method to detect task-oriented queries,developed a search task suggestion method using the concept of query expansion to answer task-oriented queries.We leveraged attributes of task frames in our knowledge base,HealthAidKB,for query expansion and sub-task mining.We evaluated the results by using queries collected from the AOL query dataset.The results of the search task suggestion evaluation show higher accuracy compared to baseline systems.The second experiment assesses the effectiveness of our knowledge base for extrinsic usage.We used a use-case scenario to evaluate an extrinsic evaluation of the knowledge base in finding relevant videos from YouTube.To this end,we built a ground truth dataset from embedded YouTube videos in our knowledge base to measure the coverage of retrieved relevant videos.The result of the extrinsic evaluation against ground truth also shows better performance of our system compared to baseline systems.(3)Our procedural knowledge base,HealthAidKB is released freely for public access?.