Throughout the history, people have been accustomed to leave lifelogs, or pieces of information regarding their personal experiences. However, such records are usually written in unstructured natural languages, making a challenge for today’s applications to access. In this thesis, we propose a framework to construct a personal knowledge base automatically. We build a FrameNet semantic parser to extract interactions between semantic roles as a basis to generate the factual triples in the personal knowledge base. For evaluation, we construct a dataset from scratch and design universal annotation guidelines for lifelogs. In the experiments, we test the system performance on personal knowledgebase construction from natural language and analyze the robustness of the proposed methodology.