The main inspiration of the project comes from Too Long; Didn't Read. We wanted to build something which can convert the boring text into something fun and interactive. Visualizations have proven to be a really effective way of imparting knowledge and our project is aimed at making it fun for everyone.
We have created a pipeline of machine learning and deep learning models which processes raw text from a website for e.g. a wikipedia article, into a Concept Map with entites and relationships. This knowledge graph will be interactive and is aimed at providing users an alternative method of knowledge representation.
We are using pre-trained deep learning models to extract, normalize and link entities. Our relationsip model extracts relationships provided the entities and text. The final step is preparing triples out of the model outputs and produce an interactive graph. We have used python, java, html and css to build a chrome extension that can be easily used.
We faced a lot of initial roadblocks related to building a chrome extension and how will it communicate with our ml and dl models. Finding a robust visualization library which builds interactive graphs also took more than required effort. For building our pipeline we had to fine-tune our models. Coreference Resolution was a major problem and still not in great shape. We had to filter relationship types to remove some generic outputs.
We all are really proud of our hard work. To build a chrome extension from scratch in just few weeks was challenging but in the end it was all worth it. Although our work is not finished we are giving our best to make this project standout and build something which can help people in their daily life.
We learned how effectively work in a team. For many of us it was the first time to work on a ml/dl project but still we figured out how to work with text processing and deep learning. The open source community was of great help.
The current chrome extension is a generic tool, in future we are looking to build it for specific domains. For e.g. an extension that can process abstracts from research papers and build a graph of all the important concepts, so that people can just quickly look and see whether a paper is relevant for them or not. We are also thinking about providing a button for downloading the created graph for users to review it later. Moreover, we are also working on integrating other open source knowledge graphs like KBpedia. In addition to this we are also working on providing an option for incrementally updating the knowledge graph in a single browsing session.
Mohammad Alshaikhusain
