One of the most important concepts in NLP is one of distributed representations, inspired by the linguistic field of distributional semantics. So we needed to find an ideal middle ground between complexity and performance. We didn’t have the resources to build a full-scale bot that could recognize the user intention among thousands of possibilities, yet we wanted to create something that could be relevant enough so that people would find it useful. However, this doesn’t take at all into account all the nuances of the language. At the most simple, one could simply look for some keywords such as hello, restaurant, or visa. There are many approaches possible to make sense of what the user wants. The main challenge when building a bot is relevancy, and this starts by having a clear understanding of what the user’s intention is. Understanding the meaning behind a question The only thing remaining was to build the brain behind the messages. We chose to build our bot with Microsoft’s Bot Framework SDK for easy development, user management and to be able to easily publish it to multiple platforms like Messenger or Line, the most popular messaging platform in Taiwan. ![]() ⚡️ Lightweight and runnable on a small server.So we looked for the best way to build a system that would be: We also didn’t want to spend a lot of time to compile a large training dataset. However, being just a small team of 3 doing this in our spare time, we didn’t have enough resources and time to build something very sophisticated. When it comes to chatbots, there are a lot of ways to go, and many tools and libraries out there to help you make your plan a reality. We plan to expand the bot capabilities in future versions. Specifically, we chose to focus on visa issues and the recently created Gold Card program. ![]() We decided to start with a limited scope first and to focus on answering practical questions about moving to and living in Taiwan. We think that a fun and approachable chatbot could help people understand a lot more about all the great things this place has to offer, as well as answer most of the questions they might have about living here. The projectĭespite being an amazing place to live, Taiwan is still misunderstood by most foreigners. There is even a Lite version of the model, small enough to run in Javascript on the client-side. Moreover, it can run much faster than BERT or other similar Transformer models and is thus more applicable to real-world problems. It is thus ideal for transfer learning and performs competitively with more complex models like BERT. Trained in a multi-tasking fashion, the model can encode sentences into meaningful continuous representations that work well on a range of different tasks. The Universal Sentence Encoder, recently released by Google AI, is one of these new models available via Tensorflow Hub. Moreover, the democratization and open-source sharing of cutting-edge deep learning models from research work at large tech companies like Google or Facebook is making it possible for anyone to implement the latest state-of-the-art solutions. ![]() ![]() Fortunately, there has been impressive progress in the fields of machine learning and natural language processing (NLP) in the past few years. The idea is to build a go-to assistant to help foreigners answer their questions about moving to, working, and living in Taiwan (pro-tip: ask the bot where to find cheese or chocolate).īuilding a functional and useful chatbot is a non-trivial project. Currently living in Taiwan, I recently joined the Taiwan Bot □ project along with Shawn and Eric.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |