TLAB at the NTCIR-13 AKG Task Md Mostafizur Rahman Sokendai(Graduate University for Advanced Studies) National Institute of Informatics Tokyo, Japan rahman@nii.ac.jp Atsuhiro Takasu National Institute of Informatics Tokyo, Japan takasu@nii.ac.jp ABSTRACT In recent years, popular search engines are utilizing the power of Knowledge Graph(KG) to provide specific answers to queries and questions in a direct way. It is expected that search engine result pages (SERPs) will provide facts about the quires satisfying semantic meaning, which encour- aging researchers to constructing more powerful Knowledge Graph. One of the major challenges is disambiguating and recognizing entities and their actions stored in KG in a con- text. To achieve and advance the technologies related to ac- tionable knowledge graph presentation, Action Mining (AM) is an essential step and relatively new research direction to nurture research on generating such KG that is optimized for facilitating entity’s actions e.g. for entity“Donald J. Trump” most potential actions could be “won the US Presidential Election” or “targeting US journalists”. This paper presents the Action Mining (AM) task organized by NTCIR-13. We employ a probabilistic model to address the AM problem. Team Name TLAB Subtasks Action Mining (AM) Keywords Entity, Actions, Action Mining 1. INTRODUCTION Data is generating from everywhere around us all the times. Smart phones, sensors and social networking sites produce tons of data everyday. In recent years, large amount of data is being available on web so handling abundance of information and extract facts can be considered as major challenges of search engines. Most of the people consider search engines as an expert of all domain. As the expecta- tion is heading towards peak, effective use of KG in SERPs becomes essential. To achieve such goals, it is mandatory to design more structured and sophisticated knowledge graph generation technique. Large Knowledge Bases (KB) e.g. DBpedia, YAGO, Deep- Dive are incorporating huge number of entities and attributes to keep pace with the high information generation in web, smart systems and social life. Evolving with new facts, at the same time organizing and maintaining existing knowl- edge is more critical. The purpose of AKG defined by NTCIR-13 1 [3] is: select and rank attributes of entities in KGs that can best support “actionable” search intents. To prepare a KG for support- ing actionable search two major steps are: (1) Recommend the actions relevant to queries, (2) Ranking the attributes of query entity based on action and graph generation. This study presents, action mining technique to foster the action- able knowledge graph generation. In short, main goal is to find the top actions relevant to query entity and entity type and embed the entities with their related counterparts. To the best of our knowledge, there is no existing work on the entity oriented action mining for actionable knowledge graph generation. Related entity mining and semantic role labelling (SRL) can be considered as the similar topics to the problem presented in this paper. In related entity rec- ommendation, goal is to retrieve top related entities given a keyword query. Web search engines more often use their own data gathered from users as well as user click logs and ses- sions to recommend related entities [4, 12, 1]. In this study, we employ mostly publicly available data to generate a list of top actions relevant to query. In SRL, predicates or verbs use for detection of the semantic arguments and identify the role of entity [5]. In AM problem based on entity we recom- mend the top related action that match with query entity and entity type. We propose a simple but statistically sound probabilistic model and discuss the parameter estimation of the model. 2. PROBLEM STATEMENT In this section, apart from the NTCIR description, we formally define the action mining problem with some exam- ples and briefly describe our approache to address the AM problem. 2.1 Problem Definition In entity oriented action mining problem, the goal is to find the potential actions and return top-k potential actions relevant to query, where query is a set of entity instance and entity type [3]. According to English grammatical rule, subject, object and verb form a sentence. The subject and the verb are the minimum requirements for constructing a basic English sentence. Verb plays the key role to give semantic meaning to a sentence. As we know that an auxiliary verb 2 is used in forming tense and a linking verb 3 joins the the subject with 1 http://research.nii.ac.jp/ntcir/ 2 https://en.wikipedia.org/wiki/Auxiliary verb 3 https://en.wikipedia.org/wiki/Linking verb 357 Proceedings of the 13th NTCIR Conference on Evaluation of Information Access Technologies, December 5-8, 2017 Tokyo Japan